RDP 9806: Policy Rules for Open Economies 5. The Perils of Inflation Targeting

This section turns from instrument rules to target rules, specifically inflation targets. In the closed-economy Svensson-Ball model, inflation targeting has good properties. In particular, the set of efficient Taylor rules is equivalent to the set of inflation-target polices with different speeds of adjustment. In an open economy, however, inflation targeting can be dangerous.

5.1 Strict Inflation Targets

As in Ball (1997), strict inflation targeting is defined as the policy that minimises the variance of inflation. When inflation deviates from its target, strict targeting eliminates the deviation as quickly as possible. I first evaluate this policy and then consider variations that allow slower adjustment.

Trivially, strict inflation targeting is an efficient policy: it minimises the weighted sum of output and inflation variances when the output weight is zero. Strict targeting puts the economy at the Northwest end of the variance frontier. In Figure 1, the frontier is cut off when the output variance reaches 15; when the frontier is extended, the end is found at an output variance of 25.8 and inflation variance of 1.0. Choosing this point implies a huge sacrifice in output stability for a small gain in inflation stability. Moving down the frontier, the output variance could be reduced to 9.7 if the inflation variance were raised to 1.1, or to 4.1 if the inflation variance were raised to 1.6. Strict inflation targeting is highly suboptimal if policy-makers put a non-negligible weight on output.

The output variance of 25.8 compares to a variance of 8.3 under strict inflation targeting in the closed-economy case. This difference arises from the different channels from policy to inflation. In a closed economy, the only channel is the one through output, which takes two periods (it takes a period for r to affect y and another period for y to affect π). With these lags, strict inflation targeting implies that policy sets expected inflation in two periods to zero. In an open economy, by contrast, policy can affect inflation in one period through the direct exchange-rate channel. When policy-makers minimise the variance of inflation, they set next period's expected inflation to zero:

Equation (9) implies large fluctuations in the exchange rate, because next period's inflation can be controlled only by this variable. Intuitively, inflation in domestic-goods prices cannot be influenced in one period, so large shifts in import prices are needed to move the average price level. (Appendix A formalises this interpretation.) The large shifts in exchange rates cause large output fluctuations through the IS curve.

This point can be illustrated with impulse response functions. Substituting Equations (5) into (9) yields the instrument rule implied by strict inflation targeting:

(Note this is a limiting case of Equation (7) in which the MCI equals the exchange rate.) Using Equations (4), (5) and (10), I derive the dynamic effects of a unit shock to the Phillips curve. Figure 2 presents the results. Inflation returns to target after one period, but the shock triggers oscillations in the exchange rate and output. The oscillations arise because the exchange rate must be shifted each period to offset the inflationary or deflationary effects of previous shifts. These results contrast to strict inflation targeting in a closed economy, where an inflationary shock produces only a one-time output loss.[5]

Figure 2: Strict Inflation Targets – Responses to an Inflation Shock
Figure 2: Strict Inflation Targets – Responses to an Inflation Shock

5.2 The Case of New Zealand

These results appear to capture real-world experiences with inflation targeting, particulary New Zealand's pioneering policy in the early 1990s. During that period, observers criticised the Reserve Bank of New Zealand for moving the exchange rate too aggressively to control inflation. For example, Dickens (1996) argues that ‘whiplashing’ of the exchange rate produced instability in output. He shows that aggregate inflation was steady because movements in import inflation offset movements in domestic-goods inflation. These outcomes are similar to the effects of inflation targeting in my model.

Recently, the Reserve Bank of New Zealand has acknowledged problems with strict inflation targeting:

If the focus of policy is limited to a fairly short horizon of around six to twelve months, the setting of the policy stance will tend to be dominated by the relatively rapid-acting direct effects of exchange rate and interest rate changes on inflation. In the early years of inflation targeting, this was, in fact, more or less the way in which policy was run… Basing the stance of policy solely on its direct impact on inflation, however, is hazardous… [I]t is possible that in some situations actions aimed at maintaining price stability in the short term could prove destabilising to activity and inflation in the medium term. [Reserve Bank of New Zealand, 1996, pp. 28–29]

The Reserve Bank of New Zealand's story is similar to mine: moving inflation to target quickly requires strong reliance on the direct exchange-rate channel, which has adverse side-effects on output.[6]

5.3 Gradual Adjustment?

The problems with strict inflation targeting have led observers to suggest a modification: policy should move inflation to its target more slowly. The Reserve Bank of New Zealand has accepted this idea; it reports that ‘in recent years the Bank's policy horizon has lengthened further into the future’ and that this means it relies more heavily on the output channel to control inflation (p. 29).

In the current model, however, it is not obvious what policy rule captures the goal of ‘lengthening the policy horizon’. One natural idea (suggested by several readers) is to target inflation two periods ahead rather than one period:

This condition is the one implied by strict inflation targeting in the closed-economy model. In that model, the condition does not produce oscillations in output.

In the current model, however, Equation (11) does not determine a unique policy rule. Since policy can control inflation period-by-period, there are multiple paths to zero inflation in two periods. By the law of iterated expectations, +1 = 0 in all periods impies +2 = 0 in all periods. Thus a strict inflation target is one policy that satisfies Equation (11). But there are other policies that return inflation to zero in two periods but not one period.[7]

The same point applies to various modifications of Equation (11). For example, in the closed-economy model, any efficient policy can be written as an inflation target with slow adjustment: +2 = qEπ+1,0 ≤ q ≤ 1. This condition is also consistent with multiple policies in the current model. There does not appear to be any simple restriction on inflation that implies a unique policy with desirable properties. Policy-makers who wish to return inflation to target over the ‘medium term’ need some additional criterion to define their rule.[8]


Black, Macklem and Rose (1997) find that strict inflation targeting produces a large output variance in simulations of the Bank of Canada's model. Their interpretation of this result is similar to mine. [5]

The Reserve Bank discusses direct inflation effects of interest rates as well as exchange rates because mortgage payments enter New Zealand's CPI. [6]

An example is a rule in which policy makes no contemporaneous response to shocks, but the exchange rate is adjusted after one period to return inflation to target in two periods. [7]

Another possible rule is partial adjustment in one period: +1 = . This condition defines a unique policy, but the variance of output is large. The condition implies the same responses to demand and exchange-rate shocks as does strict inflation targeting. These shocks have no contemporaneous effects on inflation, so policy must fully eliminate their effects in the next period, even for q > 0. [8]