9 Implementation Errors and Incomplete Information

Andrew Berg, and Rafael Portillo
Published Date:
April 2018
  • ShareShare
Show Summary Details
Rafael Portillo, Filiz Unsal, Stephen O’Connell and Catherine Pattillo 

1 Introduction

As discussed in Chapter 6, recent empirical work has failed to find much evidence of a monetary transmission mechanism in low-income countries, including in Africa. Some researchers have concluded that structural features of these economies, e.g., shallow financial markets, limit the ability of monetary policy to deliver on price and macro stability, and that other policies may be better suited to this task.

In this chapter we show that limited effects of monetary policy can reflect shortcomings of existing policy frameworks in these countries rather than (or in addition to) structural features. We focus on two issues that are pervasive, as emphasized in Chapter 1. First, central banks often lack effective frameworks for implementing policy, so that short-term interest rates display considerable volatility. Second, clear communication is often lacking, as attested by indices of policy transparency, which makes it difficult for market participants to understand policymakers’ intentions.

We introduce these features in an otherwise standard New Keynesian model. We model implementation errors as insufficient accommodation of shocks to money demand, which creates a noisy wedge between actual interest rates and the level intended by policymakers. The latter is not directly observed by the representative agent (incomplete information) and must be inferred from movements in interest rates and money.

Under these conditions we show that exogenous and persistent changes in the stance of monetary policy can have weak effects on the economy. This is the case even though the underlying transmission mechanism is strong, as reflected in the effects of the same policy under complete information. We believe our finding is important: if policy shortcomings are the source of weak transmission, then the solution is to improve on existing frameworks rather than give up on monetary policy altogether.1

2 The Model

The economy is described by a New Keynesian model, consisting as in other similar models in this book (e.g. Chapter 15) of a forward-looking IS equation and an expectations-augmented Phillips curve:

where yt is output, πt is the inflation rate, and Rt is the short-term interest rate. Output and inflation rate are determined on the basis of (t-1) information, i.e., they are predetermined. The economy also features a demand for real money balances (mt) similar to those found in other chapters (e.g., Chapter 8):

where ut is an unexpected change in money demand. It is composed of two shocks: ut = u1,t + u2,t. The role of these two shocks will become clear later. Nominal balances (Mt) can be written as:

where Pt is the nominal price level.

3 The Central Bank

3.1 Policy Intentions

Monetary policy is guided by a Taylor rule:

We refer to zt as the stance of policy, subject to shocks εt and with persistence ρ. Policy intentions can also be represented as a money rule by combining (3), (4), and (5):

3.2 Policy Implementation

The central bank implements policy by setting Mt: The objective is Mt=MtT, so that Rt=RtT. However, the central bank makes implementation errors: it accommodates u1,t but not u2,t. Monetary aggregates therefore follow:

As a result, actual interest rates differ from the intended policy stance (RtRtT):

4 Complete Versus Incomplete Information

The representative agent knows all the parameters of the model, including the specification of the interest rate rule and money demand, and the volatilities of the shocks.

Under complete information, the representative agent observes all variables (yt-1,Rt-1t-1,mt-1,Mt-1, Pt-1,), policy intentions (Rt1T,Mt1T), and shocks (u1,t-1, u2,t-1, zt-1). Under incomplete information, the representative agent observes the macro variables but does not observe the shocks nor the policy intentions. The agent observes two linear combinations of the three shocks, however. First is the difference between interest rates and the endogenous monetary policy response:

Second is the difference between money and the level implied by the endogenous determinants of the money rule in (7):

The agent faces a signal extraction problem: to infer the stance of policy, zt-1, on the basis of two noisy signals (resRt-1 and resMt-1).2

The Model Solution under Complete and Incomplete Information

Under complete information, it is straightforward to show that:

where FIt-1 = (zt-1, u1,t-1, u2t-1), and:

When a shock to monetary policy hits at time t, starting from zt-1 = 0, the immediate effect on output and inflation is zero since E[zt|FIt-1] = 0- The effect from t+1 onwards is given by yt+j = Ψyρjzt and πt = Ψπρjzt, for j = 1,2….

Regarding money demand, accommodated shocks (u1,t) have zero effects. In principle, unaccommodated shocks could affect the economy. However, since output and inflation are predetermined and shocks to money demand are i.i.d., it follows that E[u2,t+j|FIt+j-1] = 0 for j = 0,1,2…. As a result errors in policy implementation have no effects.

Under incomplete information, the solution is similar:

where LIt-1=(resRt-1, resMt-1). The estimates are derived using the Kalman filter (see Hamilton, 1994: ch. 13). Just as in the case of full information, E[u2,t+j|FIt+j-1] = 0 for j = 0,1,2…. Errors in policy implementation still have no direct effects, though they now affect the estimation of E[zt|LIt-1]. For the Kalman filter problem at time t the state equation is given by (6)

and the observation equations are given by:

Under incomplete information, when a shock to monetary policy hits at time t, starting from zt-1 = u1,t-1 = u2,t-1 =0, the immediate effect on output and inflation is also zero since E[zt|LIt-1] = 0. The effect from t+1 onwards is given by ygapt+j = ΨyρjE[zt|LIt+j-1] and πt = ΨyρjE[zt|LIt+j-1], for j = 1,2…. Unlike in the case of full information, the estimate of zt changes over time, as the persistence of the effect of zt on LIt+j leads the private sector to reassess its initial estimate.

5 Simulations

We focus on a policy loosening of 1 per cent, i.e., εt = —1, and compare the responses of output and inflation under complete and incomplete information. We use a standard calibration, summarized in Table 9.1. The Kalman filtering also requires that we calibrate the standard deviations of the shocks. Under this calibration, zt is slightly more volatile than u2,t(σz2=3.33), which implies that policy loosening (zt) and errors in policy implementation (u2,t) each account for about half of the deviation between interest rates and what is implied by the systematic part of the Taylor rule.

Table 9.1.Parameters and Standard Deviations
σ1Inter-temporal elasticity of substitution
β0.99Discount factor
η0.5Interest rate elasticity of money demand
γ1Income elasticity of money demand
κ0.34Sensitivity of inflation to output
øπ1.5Taylor rule coefficient
øy0.5Taylor rule coefficient
ρ0.7Persistence of monetary policy shock
σe21Volatility of monetary policy shock
σu123Volatility of accommodated money demand shocks
σu223Volatility of unaccommodated money demand shocks

Figure 9.1 shows the impulse responses. Under complete information, a 1 per cent decrease in the policy rate at time t raises inflation by 0.46 per cent and output by 0.42 per cent in period t+1, and with both output and inflation displaying the same persistence than the policy shock. Under incomplete information, the impact is much lower and takes longer to have an impact: inflation and output peak in period t+2 at 0.13 and 0.12 per cent, respectively.

Figure 9.1.Impulse Response Functions Under Complete and Incomplete Information

The stark difference between the two cases reflects the difficulty of inferring policy intentions when implementation errors are common. Under these circumstances policy actions have little effect on the economy. The representative agent eventually realizes the extent of policy accommodation, but by then the stance of policy has been largely corrected.

Relative volatility matters. If σu12=σu22=σe2=1, the effects on inflation and output peak at 0.22 and 0.21, respectively. Also, both shocks to money demand are needed for the result to hold. If u2 = 0, z can be inferred from resR; if u1 = 0, z can be inferred from resM. Greater persistence of z (a higher ρ) also increases the effects of policy, as the signal extraction improves over time.

6 Conclusion

We have shown that the combination of errors in policy implementation and incomplete information regarding policy intentions can greatly reduce the effects of policy on output and inflation. We believe our model can help understand the limited evidence of the monetary transmission mechanism in developing countries.

Our focus has been on shocks, consistent with the empirical literature. However, the issues here also apply to the endogenous component of policy. This can be the case if agents have less information about the state of the economy than the central banks.3


    Bindseil, U. (2004). Monetary Policy Implementation. Oxford: Oxford University Press.

    Dotsey, M. (1987). Monetary Policy, Secrecy, and Federal Funds Rate Behavior. Journal of Monetary Economics, 20, 463–74.

    Hamilton, J. (1994). Time Series Analysis. Princeton: Princeton University Press.

    Melosi, L. (2012). Signalling Effects of Monetary Policy. Working Paper 2012–05. Chicago: Federal Reserve Bank of Chicago.


This is consistent with the results of Chapter 5, which finds that a large joint policy tightening observed in East Africa in 2011 had stronger effects in those countries where the tightening was clearly communicated.

Formal analysis of incomplete information regarding central bank intentions goes back to Dotsey (1987); a recent partial equilibrium discussion is provided by Bindseil (2004) for the case of the ECB.


The representative agent also observes the difference between M and the endogenous determinants of money demand: resMt1*=Mt1γygapt1+ηRt1Pt1=u1,t1+u2,t1. But this does not provide an additional source of information, as resMt1=resMt1*+ηresRt1.

    Other Resources Citing This Publication