The Power of Open-Mouth Policies

It is a policy that works through the announcement rather than the implementation, and it needs its own method because the announcement is a dated, one-off event rather than a recurrent random draw. The authors observe that “central banks increasingly rely on communication to implement their monetary policy”, giving forward guidance and announced changes in the inflation target as examples, and note that “little work has been done on evaluating the effects of the open mouth policy within a DSGE framework.” The technical obstacle is that such announcements “can be modeled as anticipated non-Markov news shocks”, which “lead to nonstationary solutions (time-dependent optimal decision [rules]) which cannot be analyzed using conventional numerical methods for constructing one stationary (time-invariant decision function).” The distinguishing feature of the setting, as the authors put it, is that “agents react to the news of a policy change even when the new policy will not take effect until a later date.”

It constructs a path of decision functions rather than a path of variables, by analogy with the extended path method of Fair and Taylor (1983). The authors state the difference directly: “EP constructs a path for variables (time series) whereas EFP construct[s] a path for decision functions.” The approach applies to economies with the turnpike property — those “in which finite-horizon trajectories converge to infinite-horizon trajectories as the time horizon increases” — which is what allows a finite computed path to stand in for the infinite-horizon problem. The contribution here is specifically the perturbation implementation: the authors report it is “comparable in accuracy to a global projection method developed in MMTT (2020), but it is tractable in problems with much higher dimensionality, such as large-scale central-banking models.” The software is written in Dynare with a MATLAB interface.

A scaled-down replica of the Bank of Canada’s ToTEM, which the authors call “baby” ToTEM or bToTEM. The full ToTEM has 356 equations and unknowns including 215 state variables; bToTEM, developed in Lepetyuk, Maliar and Maliar (2020), has 47 equations and unknowns including 21 state variables, and the authors report it “generates realistic impulse-responses of the Canadian economy to shocks, which are very similar to those produced by the full scale ToTEM model.” It is a small open-economy New Keynesian model with three Phillips curves — from sticky domestic prices, sticky wages and sticky import prices — rule-of-thumb price setters following Galí and Gertler (1999), quadratic investment adjustment costs and convex capital-utilization costs, with international trade in consumption goods, commodities and imported inputs. Results are reported as percentage deviations from the risky steady state, which the authors define as “a state to which a stochastic economy converges in the absence of exogenous shocks”; the interest and inflation rates are in annualized percentage-point deviations.

Output rises temporarily, and it rises by more when the same change is announced a year before it starts. In the experiment the central bank announces at t = 1 a fully credible, gradual increase in the target from 2% to 3% over eight quarters. Implemented immediately, inflation tracks the target — the authors attribute this to price setters being “mainly non-optimizers who index their price by inflation target” — the nominal rate rises gradually by 1% over the first fifteen periods and stays there, and output, investment and commodity exports jump, with peak output about 0.2% above the risky steady state before descending after about a year. Delaying the same change by one year leaves the paths qualitatively similar but raises the peak in most variables, with output reaching about 0.3%; the authors attribute the difference “entirely” to anticipation, with agents accumulating capital in advance of an expected lower real interest rate. The paper’s summary of the result is that “postponing an increase in the inflation target by one year produces an additional 0.1% increase in output over the transition to a new steady state.” They report the model’s maximum equation residuals on this path range between about 0.07% and 0.0001%, and that residuals are of similar size in the remaining experiments.

Anticipation effects are still present, both before and after the uncertainty resolves. The authors model a 50% chance that the target rises gradually from t = 5 and a 50% chance it does not, implementing this by setting the period-4 expectation functions to a weighted sum over the two period-5 realizations. From t = 1 all variables show mild increases attributable to anticipation; once the outcome is known, variables return quickly to the original steady state if the target does not rise, and follow a more pronounced hump toward a new steady state if it does. An appendix varies the probability to 25% and 75%, and the authors report that in the no-change branch “the transition back to the old steady state is significantly faster for the 25-percent case than for the 75-percent case.”

Postponing the return to the Taylor rule raises the cumulative output expansion but not the impact response — so the horizon matters for the total effect while the forward guidance puzzle does not appear on impact. The economy starts at the effective lower bound; at t = 1 the bank announces the rate will be held there for T periods before reverting to the standard Taylor rule, and the authors compare T = 1, T = 4 and T = 8. On announcement the exchange rate, inflation and the real variables all jump; the currency depreciates, exports of commodities and noncommodity goods rise, and output, labor, investment and capital increase. The output expansion is largest for the eight-quarter horizon, with the peak increase “70 percent higher than the one for the forward guidance horizon of four quarters” — but “in all three cases, an initial output jump is of equal size.” The authors name the dependence of the initial reaction on the guidance horizon as the forward guidance puzzle, note it is absent here, and conclude that “the policy horizon matters for the total effect on output: it reacts more if the policy change is postponed further away in the future.” They present this as addressing two of the four normalization questions raised after the Great Recession (whether to normalize now or later, and whether to announce the shift in advance), leaving the gradual-versus-all-at-once and time-versus-state-dependent questions for future research.

No — the authors describe the total effects as “not quantitatively important”, even though the parameter change they impose is large. They double the sensitivity to inflation, the sensitivity to the output gap, or both, announced at t = 1 and implemented at t = 2, and note the change “is quite large relative to what a central bank would typically consider.” Both are inflationary. Doubling the inflation sensitivity is the more expansionary of the two, raising output, consumption, investment, capital and labor visibly at peak and in the new steady state, with a slight drop in commodity production tied to lower commodity exports; doubling the output-gap sensitivity has more modest effects; doing both gives something in between, which the authors explain by the trade-off in the rule — “responding stronger to the output gap undoes the effects of stronger responses to inflation.” The risky-steady-state interest rate is lower when the response to the inflation gap is stronger and higher when the response to the output gap is stronger, because a stronger inflation response requires a smaller rate increase to achieve the same inflation stabilization and a stronger output-gap response requires a larger one. Their summary: “switching to a significantly more aggressive Taylor rule has only minor effects on the economy’s behavior when the economy is not hit by any shocks.”

It is expansionary in the long run, but here delay is costly rather than beneficial — the opposite of the inflation-target experiment. Under price-level targeting the rule responds to the gap between the actual price level and a target price level that itself grows at the inflation target, so the bank reverses past misses rather than letting bygones be bygones; the authors note the consequence that agents “will be much more confident on where the prices will be in the future, even with a positive average inflation.” The new rule implies higher steady-state levels for all the model’s variables. Comparing immediate implementation with implementation one year after announcement, “for all of the variables (except of the nominal interest rate), the immediately implemented policy gives larger benefits”, because the economy reaches the new steady state at about the same time either way while the transition gains are smaller. The paper’s framing in the introduction is that since price-level targeting has been argued to be welfare improving, “a central bank that waits to implement the new policy in practice loses time.”

The direction survives, and the anticipation effect softens the downturn rather than amplifying it. In a variant the economy receives a permanent negative foreign-demand shock, modeled as a negative innovation to a random walk. Output and labor fall by about 1 and 1.5 percent on impact in all three scenarios considered. Without a policy switch, output recovers somewhat but its new steady state remains below the old one; with the switch, output ends up nearly at its pre-shock level and consumption is higher. The anticipated (delayed but announced) switch produces impulse responses lying between the no-change and immediate-change cases, with a smoother nominal interest rate path and correspondingly smoother behavior in the other variables. The authors note that each of the policies considered implies the central bank tightens, “even in the economic downturn.”

Very modest anticipation effects, which the authors explain by its position between the two other regimes. The rule replaces current inflation in the Taylor rule with an average of current and past inflation over M lags; the authors set M = 8, the largest value in the optimal range of 2 to 8 that Amano et al. (2020) find. They describe average inflation targeting, following Nessén and Vestin (2005), as “a middle ground between price-level targeting and inflation targeting”. The result is that “this policy change has very modest anticipation effects on the economy in the absence of any shocks”, with larger responses in output, labor, imports and noncommodity exports when it is implemented immediately — so reacting to average inflation “smooths out dynamics to a new steady state” and the authors “would not expect the economy to experience any drastic changes in the course of transition”.

Enough to change the answer: in the same experiment, a Markov news-shock treatment produces effects the authors call dramatically larger. They rerun the aggressive-Taylor-rule experiment using two adaptations of Schmitt-Grohé and Uribe’s (2012) perturbation method — one with a unit-root process for the Taylor coefficient, one with a temporary news shock — and compare. All three behave qualitatively alike, with a more aggressive central bank raising inflation, lowering nominal and real rates, and raising output, investment, capital and imports. Quantitatively the permanent Markov news-shock model predicts much larger effects for both the anticipation window and the steady-state differences, with “anticipation effects in investment … five times larger at peak”; the temporary Markov news shock returns the economy to the old steady state, while the extended function path and permanent Markov solutions converge to different new ones. The authors’ explanation is about what the decision rule is responding to: in their framework “the agent’s decision rule is the best response to a given news shock”, whereas under a Markov process the response is shaped by the assumed process as well as by the news, so “the Markov news-shock approach significantly overstates the importance of a given anticipated event because a unit-root process … implies that once a news shock happens, its effects will persist forever.” They add that the difference between the Markov and their second-order solutions comes from the slopes of the decision rules, and that under a first-order perturbation the two Markov solutions would leave the economy at the deterministic steady state while theirs would not.

As complementary rather than superior, with a clear division of labor by the kind of event. Their statement is that “the stationary Markov literature focuses on unanticipated recurrent events like business cycles, whereas the EFP method analyzes anticipated non-recurrent [historical] events like Brexit.” They are explicit that in special cases the two coincide: a one-time non-Markov parameter shift resembles a regime-switching Markov model with an absorbing state, and also a news-shock Markov model with random-walk news. Their claim is limited to the complex cases — scenarios “composed of non-recurrent periods of growth, transitions, shifts, drifts” — which they say “can’t be adequately modeled by using a single stationary Markov process”, while allowing that stitching together a sequence of stationary Markov solutions might approximate them, with tractability unknown.

It depends on the experiment, and the paper resists a single headline number. The authors’ summary is that “the model’s implications about the importance of anticipation effects depend on a specific experiment considered: there are substantial policy anticipation effects present in our experiments (1)–(2), but such effects are relatively small in experiments (3)–(5).” In the conclusion this becomes: anticipation effects are strongest for policy-rate normalization after a lower-bound episode and for a gradual change in the inflation target level, and more modest for a switch to a more aggressive rule, to price-level targeting, or to average inflation targeting. The qualifier that travels with the modest results is that they are obtained “in the absence of any shocks” — the price-level-targeting experiment with a negative demand shock shows the anticipated switch doing visible work.

The results are model-specific, credibility is assumed rather than tested, and several experiments are run on an economy at rest. Every number comes from one calibrated small open-economy model of Canada, so the magnitudes are properties of bToTEM rather than estimates from data. Full credibility of the announcement is assumed in the inflation-target experiment, and the authors flag credibility as the pivot elsewhere: in discussing the Fed’s post-pandemic guidance they write that their analysis “plays up the importance of commitment to the announced policy on which hinges the desired monetary expansion.” The Taylor-rule, price-level-targeting and average-inflation-targeting experiments are conducted with no shocks hitting the economy, which the authors state as a condition on the conclusion. The comparison with Schmitt-Grohé and Uribe (2012) is a comparison with the authors’ own adaptations of that method, since as they note that paper “does not specify how their perturbation method can be used for analyzing non-Markov anticipated shocks.” Finally, the authors present the method as general — applicable to announced tax, tariff, minimum-wage and pension changes, or to events like Brexit or an election outcome — but that generality is argued rather than demonstrated here.