Key findings
- We develop dynamic factor models to forecast 3 important economic variables: business investment, residential investment, and house prices. Accurate and timely forecasts of these variables are important for appropriately setting monetary policy.
- Dynamic factor models provide a framework for incorporating various types of economic data into the policy process and provide a robust complement to human judgement in the forecast process.
- We find these models are a useful addition to the suite of tools we use to forecast the New Zealand economy, consistently outperforming simple benchmarks. Their forecast performance is broadly comparable to our Monetary Policy Statement (MPS) forecasts, although their performance was worse during the COVID-19 pandemic when data volatility was high.
- We decompose our dynamic factor model forecasts to show which data sources are driving changes in the forecasts. This can inform judgement on how much signal to take from different data types and support communication to the Monetary Policy Committee (MPC) and the public.
Why did we do this research?
Accurately assessing the current and future state of the economy is crucial to formulating monetary policy effectively. Real-time forecasting is made more difficult by the lagged and infrequent release of most national statistics.
In New Zealand, key macroeconomic variables such as gross domestic product (GDP), are only available quarterly and with a delay of around 3 months after the quarter has finished. This data can also be substantially revised in quarters after it is first released, clouding its initial usefulness to policymakers. High frequency data sources can assist in monitoring the state of the economy in real time and forecasting future changes. However, higher volumes of data also lead to new challenges; most significantly, the need to discern between signal and noise across the many data sources, to collate a consolidated and consistent view on the state of the economy.
Dynamic factor models are particularly useful for extracting information from large datasets with both official statistics and high frequency New Zealand data, and understand which input data is the most important in determining the forecasts.
What data did we use?
The three dynamic factor models we construct use a wide range of official statistics and high frequency data for New Zealand. The types of variables we use are summarised in the table below.