Objectives of this Bulletin
This Bulletin evaluates our forecast performance in the following ways:
- Forecast accuracy — have our recent forecasts been more or less accurate than:
i. the big 4 private banks (ANZ, ASB, BNZ, and Westpac),
ii. our historic performance, and
iii. a simple forecasting model. - Forecast bias — have our forecasts been persistently too high or too low?
- Forecast error correlations —are our forecast errors correlated with errors in other variables? For example, do forecast errors for unemployment correspond with forecast errors for inflation?
Key findings
- Our forecast errors increased significantly during the COVID-19 period between February 2020 and August 2022, primarily due to large and unusual shocks. Large forecast errors were also experienced by private forecasters and other central banks during that period. Since November 2022, our forecasting performance has improved to near pre-COVID-19 levels.
- From November 2022, the Monetary Policy Committee (MPC) forecast inflation to return to the 1 to 3% target band in September 2024. Inflation did return to the MPC’s target band in September 2024 and has remained there since.
- Forecasts for non-tradables inflation have been too low for most of the period since 2020. An underestimation of domestic capacity pressures likely contributed to this. It is probable that the supply-side effects of the COVID-19 shock were particularly difficult for our forecasting models and judgements to capture.
- Forecast errors for tradables inflation have been large through most of our dataset. However, tradables inflation forecasts have not been statistically biased — that is, they have not been persistently too high or too low. A greater share of headline inflation forecast errors are accounted for by tradables, than by non-tradables, forecast errors.
- Our forecasts of GDP have been unbiased 1-quarter ahead relative to the first publication of GDP data and have been unbiased at longer horizons relative to revised GDP data. GDP revisions in the official data have been upwardly biased. GDP data typically take several years to stabilise, which supports our practice of putting more weight on labour market and business survey data to estimate the contemporaneous output gap (the difference between output and the economy’s potential output).
- Our unemployment rate forecasts have been more accurate and unbiased than other forecast variables during most of the sample period, with the exception of the COVID-19 period.
- Forecasting performance of the exchange rate has improved since we started to assume it is a random walk with no drift (i.e., a ‘flat line’). Forecasts between 2003 and 2018 tended to assume the exchange rate would revert to its historic lower level. Forecasts since November 2019 have assumed a flat exchange rate. These have been more accurate than our earlier exchange rate forecasts, and more accurate than private bank exchange rate forecasts over the same period.
- We find that our overall forecasting framework is fit for the purpose of assisting MPC to achieve low and stable inflation under most circumstances. Forecast error analysis helps us to make continued incremental refinements to our forecasting techniques. However, we find that forecast accuracy deteriorates when shocks are very large, persistent, and have not been observed in New Zealand’s modern economic data which generally begins in the late 1980s. Such shocks require more risk-based frameworks for monetary policy decision making, and a willingness to change course as the situation develops.