Some comments to Orphanides and Svensson
"The Role of Monetary Policy Rules in Inflation Targeting Regimes - Theory meets Practice"
Workshop, Norges Bank, 5 - 6 May 2003
Jan F. Qvigstad
Flexible inflation targeting requires forecasting. The inflation forecast can be seen as the intermediate target for monetary policy.
Forecasting inflation is not a mechanical process. Among the challenges faced are:
- Imperfect models that require a considerable degree of judgement
- Lack of data and data revisions
- Choice of assumptions that in some cases are critical for the numerical forecasts
In my comment today, I would like to discuss the two latter challenges: data uncertainty and the choice of assumptions.
No central bank will formulate monetary policy based on a simple instrument rule alone, but such rules may be useful as cross-checks for the monetary policy stance. Simple rules cannot be used to "fine-tune" the instrument rate. However, in a hypothetical case where the inflation target is substantially off track, simple instrument rules may provide an important "warning signal". The value of the rules as cross-checks depends on the extent to which their input can be measured with a satisfactory degree of certainty ("rubbish in = rubbish out"). As pointed out by Orphanides, measures of capacity utilisation are particularly uncertain.
Orphanides suggests using a natural growth rule' to avoid some of the measurement problems. The natural growth rule is robust to misconceptions about the natural rate of interest and less sensitive to potential output measurement problems. A rule of this type may work well in contra-factual model simulations, assuming that you in fact stick to it all the time. However, a rule that aims at specifying the appropriate change in the instrument rate is less suited to being a cross-check. It does not give you information about whether the current level of the interest rate is roughly right. Thus, it may be a good rule' for monetary policy, but not a good (final) cross-check on the suggested interest rate based on an overall assessment. And it is as a cross-check we use simple instrument rules. Still, it may be a useful device for evaluating whether historical interest rate setting has been appropriate (along with other evaluation methods, such as checking whether the target variables have been sufficiently close to their targets).
Theoretically, the output gap is a good measure of the cyclical position. But the real- time properties of GDP data are not very good. Chart 4 shows one example of output gap estimates for Norway, before and after the latest main revision of national accounts figures in May last year. The difference is striking. Used in a Taylor rule, the messages following from these two series are very different.
The problem of substantial uncertainty surrounding output-gap measures goes beyond the use of simple monetary policy rules like the Taylor rule. Uncertainty about the output gap mirrors uncertainty about the actual state of the economy. Inflation targeting requires forecasts of future inflation and other key economic variables. If the starting point, i.e. the present state of the economy, is unknown, or at least highly uncertain, making accurate forecasts becomes a very challenging exercise.
There are data with better real-time properties than GDP figures. Registered unemployment is one candidate. It is a full count of the population and available only a few days after the end of each month. In calculating the unemployment gap, we are still faced with the difficulties in measuring the natural rate of unemployment (or NAIRU, NAWRU). Chart 5 shows that our measure of the unemployment gap tracks the revised output gap (which is unknown in real time) very well.
Wage growth and credit growth are alternatives. These data are known with greater accuracy than national accounts figures. Wage growth is a measure of tightness in the labour market. There is a close relationship between activity and domestic credit growth.
Even though there are a lot of problems in "getting it right", we cannot avoid making an assessment of the output gap, mainly based on judgement. We make this assessment, and we publish it.
Let me now turn to the issue of how to choose the assumptions underlying the inflation forecast. This subject is more related to the paper presented by Lars Svensson yesterday. However, I would like to take the opportunity to raise a few questions on this issue as well, since we have the privilege of having a group of international experts to share their views with us on this subject today.
Some advice for the conduct of flexible inflation targeting in practice have emerged from the theoretical literature, where Lars Svensson has been one of the main contributors1. It is claimed that without further specification, the precise monetary policy objectives under inflation targeting are still open to interpretation and suffer from lack of transparency. The suggested solution is to publish the central bank's loss function, explicitly showing the weight put on output stabilisation relative to inflation stabilisation. In addition, central banks should publish their forecasts of inflation and the output gap. Most central banks do this, at least for inflation. However, many academics are sceptical to the use of constant interest rate paths as an underlying assumtion when making forecasts. Forecasts of inflation and the output gap, they argue, should be constructed on the basis of an optimal path for the interest rate. Today, only a few central banks base their forecasts on such optimal or "endogenous" interest rate paths.
Norges Bank's current practice is very similar to that of other inflation targeting central banks like the Bank of England and the Swedish Riksbank. Although our practice seemingly differs somewhat from the theoretical recommendations, it is in my view not so different in practice:
- We have a general targeting rule.
- Norges Bank sets the interest rate so that future inflation, normally two years ahead, will be equal to the inflation target of 2½ per cent.
- No explicit loss function is formulated or published. However, the relative weight on output stabilisation is implicit in the choice of the target horizon of (normally) 8 quarters.
- The baseline inflation forecast is normally based on constant paths for the interest rate and the exchange rate. However, the baseline scenario is frequently supplemented by one or two alternative scenarios, often based on market expectations of the interest rate and the exchange rate.
- The assumptions on interest rate and exchange rate are of a "technical" nature. Conditional on these assumptions, the inflation forecast may differ from the objective at the target horizon of two years.
- The published assumption concerning the interest rate may differ from the path we see as most likely. To guide market expectations and the public, Norges Bank uses a standardised bias-formulation. This bias signals the most likely direction of the monetary policy stance to the public.
In our last Inflation Report, we presented two alternative inflation projections, based on different assumptions about the interest rate and the effective exchange rate. Assumptions in the baseline scenario are shown in chart 8.a, the assumptions in the alternative scenario are shown in chart 8.b.
Charts 9.a and 9.b shows the corresponding paths for inflation and the output gap in the two scenarios (the chart shows the inflation gap, meaning that the inflation target of 2½ per cent is set equal to zero here). None of the projections hit the inflation target of 2.5 per cent at the 8 quarter horizon.
In the baseline scenario (solid line), based on a constant interest rate and exchange rate, inflation was below target with a negative output gap. The alternative scenario was based on market expectations of the interest rate and the forward exchange rate (according to UIP). In this scenario, the interest rate was lower than in the baseline scenario. With these assumptions, inflation was projected to be higher than the inflation target two years ahead, and the output gap remained positive throughout the projection period. The inflation target was not met in any of the scenarios. Hence, one may claim these paths gave little guidance to the public about the future stance of monetary policy.
However, the message to the markets could be interpreted as: We think the most likely path of the interest rate lies somewhere between these two assumptions. This seemed to be well understood by market participants.
The question is still: Would the use of time-varying or "endogenous" paths for the interest rate and the exchange rate in the Inflation Report have given better guidance to the public's interest rate expectations?
Alternatively, we could have presented something like the paths shown in red in chart 10. With an UIP-assumption for the exchange rate, the thick interest rate line shows one path for the interest rate that would result in inflation two years ahead equal to the target of 2½ per cent, gives the outlook described in the report.
This is not an "optimal" path in the sense that it does not follow from minimisation of a loss function. It is neither the only path resulting in an inflation rate at target two years ahead. It may still serve as a useful illustration.
The "endogenous" interest rate path is conditional on a wide set of assumptions concerning the rest of the economy. Among these is the exchange rate assumption. In my simple illustration, I have assumed that the UIP-condition holds. This gives consistency between the interest rate and the exchange rate paths.
With "endogenous" paths for the interest rate and the exchange rate, more information is given to public at the time of publication. Although the information content is qualitatively the same , the "endogenous" interest rate path is explicit about the Central Bank's plan for the future. However, the published plan for the interest rate is still contingent on the other assumptions underlying the analysis.
Following the publication of the Inflation Report, the exchange rate turned out to depreciate more sharply than in any of the two scenarios. The depreciation was also much more pronounced than in the hypothetical "endogenous" case presented here.
Svensson (2001) states that:2
"Since the constant interest rate is not the best forecast for the actual interest rate, the corresponding inflation and output gap forecasts are not the best forecasts of actual outcomes. This makes it more difficult and less relevant to compare actual outcomes to central bank forecasts."
In reality, time-varying optimal paths cannot solve this problem alone. In the discrete-time models commonly used in the literature, the central bank can publish inflation and output forecasts each period (i.e. quarter), and market participants and the public always have access to an updated "inflation report". In practice, agents live in continuous time, and the central bank's forecasts may only be available three or four times per year. If important "news" arrives, such as a large depreciation, these forecasts may become outdated very quickly. Most of the time, updated forecasts from the central bank are not available to market participants and the public . Irrespective of the choice of assumptions - constant or endogenous' - market participants have to make their own forecasts and guess how the central bank will respond to the "news".
I will argue that it is as least as important that the central bank communicates as clearly as possible how it responds to "news", that is, its reaction function, in order to provide guidance to the market participants when forming expectations about the future monetary policy stance. This would also include communicating the Bank's estimate of the monetary policy transmission mechanism. In practice, I believe this is more important for efficient communication than the choice of assumptions.
Norges Bank has in the past published estimated effects of interest rate and exchange rate movements on inflation. This spring, by comparing these estimates to the scenarios in the Report, market participants were able to get an idea of the likely consequences of the depreciation for the interest rate.
Summing up, I think that our practice is not that different from the theoretical recommendations. Still, there are some potential improvements and refinements to be considered:
- Switching to a time varying (optimal) instrument rate path is one such potential improvement. I would be particularly interested in hearing the views of those of you present here today who have experience of using such time- varying interest rate paths.
- We could perhaps also develop further the analyses we prepare for the Executive Board in the strategy documents they consider every four months. In light of the recommendations in the theoretical literature, there may be scope for expanding the set of paths for inflation and output, based on different paths for the instrument rate.
- Another question is whether a "loss function" could be used more actively in our communication. After all, a loss function explains in a straightforward way the trade-off between inflation and output variability. Is this possible without the mathematics of it?
1"The Inflation Forecast and the Loss Function" Paper presented at the Goodhart Festschrift, Bank of England, November 15-16, 2001.
2Op.cit., Page 12.