New Zealand's experience with changing its inflation target and the impact on inflation expectations

ABSTRACT

We document the experience of the Reserve Bank of New Zealand in changing its inflation target, particularly the effects on inflation expectations. First, the Reserve Bank of New Zealand's dynamic stochastic general equilibrium model is used to highlight expectation formation in the transmission following a change in the inflation target. Second, a Nelson–Siegel model is used to combine a number of inflation expectation surveys into a continuous curve where expectations can be plotted as a function of the forecast horizon. Using estimates of long-run inflation expectations derived from the Nelson–Siegel model, we find that numerical changes in the inflation target result in an immediate change in inflation expectations.

KEYWORDS:

• Macroeconomics
• monetary policy
• inflation targeting
• inflation expectations
• yield curve modelling

Acknowledgements

The views presented here are our own and not necessarily those of the Reserve Bank of New Zealand. We are grateful to Leo Krippner, Gunes Kamber, Nick Sander and Jed Armstrong for assistance with this project. We are also grateful to our colleagues at the Reserve Bank for their helpful comments and the two anonymous referees for their helpful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The various vintages of the PTA can be viewed at http://www.rbnz.govt.nz/monetary-policy/policy-targets-agreements. Further details on the history of the PTA can be found in Reserve Bank of New Zealand (2000, 2007).

2. To a degree, these changes to the PTA cemented what was already being practised. Real economy considerations are always present when thinking about how to set monetary policy to achieve an inflation target. For further details on the Bank's contemporary interpretations of Clause 4(b) of the PTA, see Hunt (2004) and Bollard and Karagedikli (2006).

3. The intention to bring inflation back to a comfortable part of the target band in a specified timeframe following any deviation from target was stated publicly in Bollard (2002, 2008).

4. Hargreaves (2002) conducts simulation experiments to show how inflation variability would change if the Bank used the target band as a ‘zone of inaction’. That is, monetary policy does not respond when inflation is inside the target band. He finds that such a strategy would make breaches of the target band much more likely.

5. Kamber et al. (2015) provide a technical description of the NZSIM model that is based on a standard open economy dynamic stochastic general equilibrium. The version of NZSIM used in the forecasting process, and in this paper, also includes additional semi-structural auxiliary equations.

6. See, for example, Karagedikli and McDermott (2016), Kamber et al. (2015), Hargreaves, Kite, and Hodgetts (2006) and Razzak (1997) for discussions on modelling inflation expectations in New Zealand.

7. For a discussion of the Bank of England's core DSGE model, COMPASS, that uses rational expectations and an alternative model MAPS for different expectation formation processes, see Burgess et al. (2013). For a discussion on the Bank of Canada's TOTEM and TOTEM II models, which uses forward-looking and rule-of-thumb expectations, see Dorich et al. (2013).

8. The parameter needs to be high but less than unity under adaptive expectations. The value of 0.92 was chosen because that is the estimated value in NZSIM.

9. These experiments are conducted under the assumption that only small changes in the inflation target are considered, so that the model's structure can be assumed constant despite the change in institutional settings (i.e. the Lucas critique is not an issue). If changes to the inflation target were large enough we may see threshold effects where the private sector, in order to avoid the costs of inflation, engages in speculative activity rather than productive activity, reducing the economy's supply potential. For evidence of how inflation beyond a certain threshold reduces growth, see Khan and Senhadji (2001).

10. See Hall and McDermott (2009, 2011, 2015) for evidence on the New Zealand business cycle. The average duration of the nine recessions since 1947 is 4.2 quarters with a standard deviation of 1.6. The last recession lasted five quarters from December 2007 to March 2009.

11. The concept of an implausibility index can seem unintuitive. An alternative way of expressing the concept follows. In this exercise, starting from equilibrium, we generate a scenario where interest rates are at the zero lower bound for five consecutive quarters using all the shocks in the model. The shocks used to generate this scenario are compared to history to inform us whether the required shocks are ‘plausible’ given what we have observed in history. This exercise is done for different inflation targets.

12. The neutral real interest rate chosen here is the same value as used in the version of NZSIM that is used to produce the forecasts in the Monetary Policy Statement. This value matches the average of a range of models used to estimate the neutral real interest rate. For details, see Richardson and Williams (2015).

13. These results do have to be treated with some caution. Exclusion of rare but large shocks from the model would alter the calculated probabilities. Research prior to the global financial crisis showed that episodes at the ZLB would be infrequent. For example, Reifschneider and Williams (2000) found that a 2% inflation target for the USA would result in monetary policy being constrained at the ZLB about 5% of the time. Recent events suggest the standard models omitted an important shock for the USA. Reinhart and Rogoff's (2009) 800-year history of financial crisis suggests that standard models are likely to be estimated over too short a period to capture episodes of banking crisis and so may underpredict the likelihood of being constrained at the ZLB. That said, Calomiris and Haber (2014, p. 454) note that Australia, Canada and New Zealand have had a crisis-free banking system since 1970. Therefore, using standard models to calculate the probability of being constrained at the ZLB may be more appropriate in these countries.

14. For studies on the benefits of model averaging and combing data, see, for example, Stock and Watson (2002) and Wright (2009).

15. See Diebold and Rudebusch (2013) for a discussion of the use of Nelson–Sigel models for modelling bond yields as well as for the derivation of the models.

16. See Lewis (2015) for a detailed description of the estimation method.

17. Due to some survey data being unavailable, we are unable to analyse the earlier inflation targeting period. Unfortunately, this means we cannot formally test whether the 1996 change to the PTA altered inflation expectations.

18. Bai and Ng (2006) show that the asymptotic distribution of the OLS estimator of a regression that contains a generated regressor from a factor model, such as the one we use, is unaffected by the estimation of the factors so long as the square root of the time dimension divided by the cross-section tends to zero. Given that in the current application, the time dimension is much larger than cross-section dimension we can safely ignore the generated-regressor problem.

19. For a study on the role of expectations in the determination of inflation in New Zealand see Karagedikli and McDermott (2016).