2 Modelling approach
2.1 Scope of the structural VAR model
The minimum set of variables to include in the structural VAR model is governed by the aim of determining the main shocks to impact on New Zealand real GDP, domestic demand, exports and prices. The choice of variables is also influenced by insights from prior research. No model can include all principal interactions in the New Zealand economy. Even the relatively large models developed by Szeto (2002) and by Black, Cassino, Drew, Hanson, Hunt, Rose and Scott (1997) omit some variables. Nevertheless, to achieve the objectives set for this paper, for a small open economy like New Zealand the relevant set of variables should include measures of the following foreign and domestic variables.
The foreign variables should include measures of foreign real output, foreign nominal interest rates and foreign real asset returns. The differing commodity make-up of New Zealand exports and imports suggests the separate inclusion of the foreign currency prices of New Zealand exports and imports, rather than the portmanteau terms of trade variable. The reason is that fluctuations in the terms of trade arise from changes in either the prices of exports or the prices of imports and their impact on the domestic economy can differ.
The domestic variables should include measures of real exports, real aggregate domestic demand, real domestic aggregate output, domestic consumer prices, a monetary policy instrument, the nominal exchange rate and domestic real asset returns. The relative importance of agricultural production in New Zealand, its sensitivity to climatic conditions and the sensitivity of production in several other industries to climatic conditions, such as primary food manufacturing and electricity generation, suggest that a measure of climatic conditions should also be included.
The basic model therefore includes 13 variables. Each variable is explained by a structural equation that has an error term associated with it. The error term for each equation is interpreted as representing a particular innovation or shock. These shocks are labelled according to the structural equation from which they derive. For example, the error term derived from the equation for export prices is given a name such as ‘export price shock’. Appropriate specification and estimation of the system of 13 equations captures the systematic effect of export prices and other relevant variables in the model on the behaviour of the domestic variables such as real exports, real domestic demand and output. The full system of equations can be used to simulate the reaction of endogenous variables such as real exports, real domestic demand and output in response to ‘export price shocks’. The system can also be used to decompose and account for past influences on macroeconomic fluctuations and the business cycle, including the influence of fiscal and monetary policy.
For simulations and for historical decompositions to have meaningful interpretations, the ability to identify the relative size and dynamic impact of shocks, such as an export price shock, depends on the variables included in the model and on the restrictions placed on each structural equation. For example, the inclusion of a variable to capture climatic conditions will facilitate the identification of the size and dynamic impact of climate shocks. If climatic conditions prove to be a significant explanation for deviations of real GDP from trend, we could infer from this result that models for New Zealand real GDP that do not account for the influence of climate would at best incorporate its effect in the error term for real GDP and perhaps the error terms for other variables and could therefore easily result in misinterpretation of the explanation for these errors or shocks.
Although 13 variables is a relatively large number to include in a structural VAR model, with the specific variables included in this model sufficient to identify particular trade, financial and climatic shocks, it is not sufficient to analyse influences that are not identifiable from the shocks that have been distinguished. It would not be appropriate, for example, to use shocks identified from the structure of this model to attempt to identify the impact of fiscal policy initiatives such as the “Think Big” capital expenditure programme of the early 1980s, the benefit reforms of the early 1990s, or the income tax changes introduced in 2000. In this model, the consequences of these fiscal policy initiatives are likely to be subsumed into the aggregate domestic demand variable and perhaps also into the aggregate domestic supply variable if they involve changes in the economy’s productive infrastructure.
One of the reasons for developing a structural VAR model to evaluate the impact of trade, financial and climate shocks is to provide a basis for eventually developing a richer model that will enable the identification and the simulation of the impact of monetary policy and fiscal policy initiatives on the domestic macroeconomy. An evaluation of the impact of monetary policy on the business cycle requires appropriate separation of monetary policy influences from all the endogenous variables in the model. This procedure and results are presented in Buckle, Kim, and McLellan (2003).
The application of the model for fiscal policy analysis has been deferred until more appropriate time series data for government transactions that separately identify revenue and expenditure flows are available. Theory and international empirical research (see for instance Blanchard and Perotti, 1999) would suggest this separation is important both for identifying discretionary fiscal policy and for understanding its dynamic impact. We considered using the net cash flow from operations variable (NCFO) derived in Buckle, Kim and Tam (2001) in their evaluation of the sources of shocks to the government budget balance. However, the NCFO variable is a net cash flow variable and, while suitable for evaluating the impact of non-fiscal shocks on the governments net cash transactions from operations, the expenditure component of that series does not include government capital expenditure.
The focus of this paper is the identification of shocks that push the New Zealand economy temporarily away from its long-run growth path. In other words, in line with the structural VAR models of Sims (1980), Bernanke and Blinder (1992) and Dungey and Pagan (2000), departures from trend are viewed as transient. This approach means the dynamics generated by each shock are directly comparable with the dynamics arising from temporary shocks simulated by the two operational calibrated New Zealand economy-wide models (Szeto, 2002; Black, Cassino, Drew, Hanson, Hunt, Rose and Scott, 1997). These models conceive of variables growing along a steady state growth path. These calibrated models are capable of changing the steady-state growth path. Although we view this as an important direction for future development, this is not a feature of the structural VAR model developed in this paper. The comparison with the calibrated models is therefore restricted to deviations around given long-run growth paths.
To make this approach operational we first detrend the variables in our structural VAR model. The appropriate choice of detrending procedure is not straightforward. A linear deterministic trend has been the preferred method for several recent Australian structural VAR models (Dungey and Pagan, 2000; Dungey and Fry, 2000; Fry, 2001; Joiner, 2002). This procedure is appropriate if the aim is to avoid removing the stochastic trend component from the growth cycle. A potential drawback of a linear trend is it can result in periods where the deviation from trend is large or is sustained for long periods of time if the underlying trend involves a significant change during the sample period. This is a feature of many of the time series included in our model. A linear trend applied to New Zealand real GDP, for example, over the sample period used to estimate this model results in historically very large output gaps during the 1980s.
The approach used in this paper is to detrend all variables using the Hodrick-Prescott filter (Hodrick and Prescott, 1997) with the exception of the climate variable, which has been detrended by removing the long run average for each quarter. These detrending procedures are consistent with the focus of the paper, which is to understand the dynamic impact of temporary shocks around a long-run growth path. Shocks may be persistent, but they are transient. The model could nevertheless be extended in the future to capture the stochastic trend component by modelling the quarterly growth rates and applying an error correction procedure.
