Calm after the storm?: Supply-side contributions to New Zealand’s GDP volatility decline,” New Zealand Economic Papers 37, forthcoming

The views expressed in this Working Paper are those of the author(s) and do not necessarily reflect the views of the Treasury. The paper is presented not as policy, but with a view to inform and stimulate wider debate. Abstract The variance of New Zealand's real GDP has declined since the mid-1980s. To investigate why, this paper decomposes the variance of chain-weighted estimates of production-based real GDP growth into sector shares, sector growth rate variances and co-variances. The principal explanation for the decline in GDP volatility is a fall in the sum of sector variances driven by a decline in the Services and Manufacturing sector production growth variances. Sector co-variances have had a dominant influence on the profile of GDP volatility and this influence has not diminished. Despite marked changes in sector shares, notably increases in Services and Primary sector shares and a decrease in the share of Manufacturing, this has not been a significant factor influencing the decline in GDP volatility. We postulate that policy interventions such as " Think Big " , regulatory interventions during the early 1980s, and the introduction of GST are key explanations for the higher volatility until the mid 1980s. Cessation of these interventions, deregulation and possibly changes in inventory management methods are important reasons why GDP volatility has fallen since then.


Background and motivation
There is mounting international evidence that growth and business cycles in many developed economies have been through a transition to lower volatility. There are also some indications that the variance of New Zealand's real GDP growth may have declined after the mid 1980s (Hall, Kim andBuckle, 1998 andReserve Bank of New Zealand, 2000). The purpose of this paper is to examine whether the volatility of New Zealand real GDP has changed and to uncover the reasons for any change.
Identifying whether real GDP volatility has changed and understanding why it has changed are important for several reasons. From a research perspective, linear models for output growth over periods that span a break in volatility may be misspecified. From a policy perspective, a reduction in the variance of output fluctuations should alter the interpretation policy makers place on a particular realization of quarterly GDP growth. What may have been viewed as a moderate fluctuation in activity prior to the break may now be viewed as more severe.
It is also important to understand whether a decline in GDP volatility is a permanent or transient feature of an economy. It may be the consequence of permanent structural changes in which case it may be enduring, or it may reflect the fact that an economy has been subjected to less severe exogenous shocks. Structural changes that could influence GDP volatility are changes in the way policy responds to and influences economic activity or in the way the microstructure of the economy responds to other exogenous shocks.
There is a growing body of international research testing for evidence of change in the volatility of economic growth and business cycles in many developed economies. McConnell and Perez-Quiros (2000), Kim and Nelson (1999) and Shaghil, Levin and Wilson (2001) provide compelling evidence of a structural break in the variance of U.S. real GDP growth in the first quarter of 1984. The extent of this change is highlighted by McConnell, Mosser and Perez-Quiros (1999) who observe that "in the period since 1984, the volatility of quarterly real GDP growth has been only half that of the preceding twenty-five years" (page 1). The variances of GDP for Canada, the United Kingdom and Australia evidently also declined sharply in the mid 1980s (see Blanchard andSimon, 2001 andSimon, 2001).
Although the recent international literature has focused on the mid 1980s as a possible turning point in the transition to less volatile GDP, Blanchard and Simon argue that this period is just one stage of a decline in GDP volatility that has been taking place over several decades. They conclude that, in the G-7 countries other than Japan, output volatility since the 1950s is characterised by a steady downward trend that was interrupted in the 1970s and 1980s when inflation was high, and returned to trend in the late 1980s and 1990s.* This feature also seems to be evident in the long-run behaviour of Australian real GDP (Simon, 2001 andGruen andStevens, 2000).
This debate is reminiscent of an earlier debate concerning the stabilization of the post-war U.S economy relative to the pre-war period (see for instance Burns, 1960;Bailey, 1978;de Long and Summers, 1986). The claims made in that debate were questioned by Romer (1986aRomer ( , 1986bRomer ( , 1991 and Watson (1994) who doubted the comparability of the data between the two periods and the reliability of pre-war reference dates relative to their post-war counterparts. The quality and comparability of GDP data over time may yet prove to be part of the explanation for the apparent long-term decline in the volatility of GDP in many G-7 countries. However, they seem less likely to be as serious in the identification of a structural break observed in many countries during the mid 1980s. Furthermore the procedures adopted in the current debate tend to allow the data, rather than an unrelated historical event, to determine the timing of any break in volatility.
There has also been interest in understanding why real GDP volatility has declined in these countries since the mid 1980s. Blanchard and Simon (2001) have examined the contribution of changes in the volatility of components of demand to changes in U.S. real GDP volatility. They conclude that the volatility of government spending was very high during the Korean War years, fell rapidly during the 1950s, and has remained low ever since; that most of the trend decline in GDP volatility can be traced to a decline in the volatility of private consumption and investment demand throughout the period since the 1950s; and the sharp decline since the mid 1980s is associated with the decline in inflation volatility.
Changes in the volatility of durables sales and inventory investment have also been identified as key explanations for the sudden decline in U.S. GDP volatility by Kahn, McConnell, Perez-Quiros (1999), McConnell, Mosser and Perez-Quiros (1999), McConnell and Perez-Quiros (2000 and Wamock and Wamock (2000). Improvements in inventory management techniques have therefore been identified as a potentially important cause of the decline in GDP volatility. These techniques include 'just-in-time' inventory management that has allowed firms to operate on lower inventory volumes thereby reducing the durables share of GDP and their contribution to volatility.
There is also a widely held view that many developed economies have learned how to use discretionary macroeconomic policies in a way that reduces instability. In particular, a view has emerged that monetary policy should be the predominant stabilization tool for'developed economies, whereas the focus of interest for fiscal policy should be issues of sustainability and inter-temporal equity rather than short-run stabilization (see for example Taylor, 2000b andAllsopp andVines, 2000). This view has been reinforced by a growing body of literature evaluating the extent to which changes in the conduct of monetary policy could have resulted in a more stabilizing influence on real GDP (see for example Clarida, Gali and Gertler, 1999, and Taylor, 2000b. For example, Taylor (2000a) considers that Federal Reserve monetary policy since the 1970s is the most probable cause of the decline in GDP volatility in the U.S.
Other suggestions include fewer and smaller supply side shocks (Taylor, 2000a;Blanchard and Simon, 2001;Simon, 2001), deregulation, globalisation and the gradual breakdown in trade barriers enabling firms to minimize risk and achieve steadier growth by exporting their products to a broader range of countries (McConnell, Mosser and Perez-Quiros, 1999), and the shift to a more serviceoriented economy that is less cyclically volatile (Taylor, 2000aand Wamock and Wamock, 2000).2,3 Zarnowitz (1999, on the other hand, takes a sceptical view of the relevance of many of these suggested explanations.

Methodology and data
Several methods have been used in the international literature to measure and identify reasons for change in the volatility of real GDP. These methods include: (i) applying a Markov switching process to document the timing of a change in total GDP volatility and to determine whether any change is due to a narrowing of the high and low growth rates or the variance of growth rates (McConnell and Perez-Quiros, 2000;Kim and Nelson, 1999); (ii) examining the times series properties of aggregate real GDP and to attempt to identify whether there have been changes to the size of exogenous shocks or the dynamic reactions to shocks (Blanchard and Simon, 2001;Simon, 2001); (iii) examining the behaviour of volatility at different frequency domains (Shaghil, Levin and Wilson (2001); and (iv) examining some of the components of the demand for real GDP, such as private consumption and investment demand, durables and non-durables demand, etc (McConnell, Mosser and Perez-Quiros, 1999;McConnell and Perez-Quiros, 2000;and Wamock and Wamock, 2000;Blanchard and Simon, 2001). The method used in this paper is similar to the fourth approach of decomposing the components of real GDP. However, the limited availability of suitable demand side data and the likelihood that there have been significant changes to the supply side structure of the New Zealand economy, especially since the mid 1980s, suggests that a different decomposition is warranted. We therefore focus on the supply side of the economy and identify the relative contributions of changes in production sector shares, variances and co-variances.
Our approach is to start by recognising that changes in aggregate GDP growth can occur as a result of changes in production sector shares of real GDP (that is, changes in their relative size) and .changes in production sector growth rates. From this perspective, changes in the volatility of aggregate GDP growth can occur as a result of any one of three proximate causes: changes in sector shares, changes in the volatility of growth of any one sector, and changes in the correlation between the growth rates of all sectors. We label these three components as follows: sector shares, sector variances, and sector co-variances. To identify the relative importance of each of these three components, we rearrange industry output series into ñve production sectors: Services, Government and community services, Manufacturing, Primary and Construction.
To avoid the 'Romer and Watson critiques', an inquiry of this nature requires data prepared on a consistent basis to avoid changes in data compilation methods influencing the interpretation of the performance of GDP and each sector. The frequency and industry availability of real GDP data are also important considerations. There are several alternative New Zealand constant price GDP data series available. They vary by the time period they span, by industry composition and by frequency of coverage. There are also significant differences in the methods used to construct indices of real GDP.
Statistics New Zealand (1998) has prepared quarterly real output data for 23 industries on an SNA basis since the September quarter 1977September quarter (1977. Until recently, these series were derived on a fixed weights basis. Statistics New Zealand The new SNA 93 chain series are now New Zealand's official data series and replace the previous official series, which were based on SNA 68, New Zealand Standard Industrial Classification (NZSIC), and fixed weights. The new chain series for production industries are available from the June quarter of 1987 onwards. The fixed series for production industries are available from September 1977. A calibrated quarterly chain linked series for real GDP and its production sector components was therefore constructed to provide a longer time span, one that spanned the period prior to the mid 1980s when other countries were evidently experiencing a decline in volatility. The construction of these production data series is described in Haugh (2001).
The calibrated series are SNZ's new quarterly chain series from June 1987 onwards appended to a calibrated chain series for the period back to September 1977. The latter-is derived by exploiting the statistical relationship between the period of overlapping -chain and fixed series (1987:2 to 2000:2). This statistical relationship is then used to derive series for each SNA93 based ANSIC production sector and for total real GDP from 1977:3 to 1987:1, which are intended to approximate the chain-linked series for this period. These calibrated series are then combined with the respective 1987:2 to 2003:2 chain-linked series available from SNZ to form consistent time-series data for each industry over the period 1977:3 to 2003:2. 4 The calibrated chain series for real GDP therefore provides a conceptually consistent database from which to understand how the New Zealand macroeconomy has been evolving over the period from 1977:3 to 2003:2. Calibrated real output series were also derived for five production sectors as well as for total real GDP. The five sectors are: Services, Government and community services, Manufacturing, Primary, and Construction. The allocation of industries to the five sectors is described in Table 1. The composition of our five sectors is easily identifiable and is consistent with United Nations SNA classifications and therefore enhances the opportunity for meaningful international comparisons.

Á decomposition of real GDP growth and its variance
The calibrated production series are used to estimate changes in the relative importance of sectors (changing sector shares), their growth rates, growth rate volatility, and the contributions of changes in sector shares, sector variances and sector co-variances to changes in quarterly real GDP volatility. To guide this analysis, it is instructive to identify the components that make up each growth and variance series.
Let Y t denote the time series of seasonally adjusted GDP levels and X pl denote its component level production sector series so where the error term E t principally accounts for the difference between seasonally adjusted GDP, Y,, and the seasonally adjusted components, X.,, but also other factors (see Haugh, 2001 for further details). Thus, and so, to a good approximation*, (2) Alogy, =£,+e, where _£_

( = 12 r
Here, e t represents the error arising from E, and, to a lesser degree, approximation error and p ¡ t is the share or proportion of GDP represented by the component X •, such that (3) Pjl =X jt /Y t ; y = l,2,...,/?;i = l,2,....,r Moreover, y, =io g y,-iog7 M represents the (continuously compounded) growth rates of Y t with similar interpretations for the A log X Jt , Typically, the share, p, t , will evolve smoothly over time by comparison with the individual growth rates A log X,,. The latter have been chosen in preference to (Xj, -X j,_, ) / Xj,_, since these are continuously compounded rates, which are the common measures of growth used in the literature. The mean and variance of A log Y t are given by where the mean and variance of S, are given by These approximate decompositions assume that the p • t are smooth by comparison to the Alog^f. t . This assumption, validated later in the paper, effectively eliminates terms involving the variability of p Jt . In practice the decompositions are calculated over moving windows within which the shares, Pi t , vary insignificantly about a stable value by comparison to the growth rates, A log*,.
The structural decomposition (2) and the mean and variance decompositions (4) to (7) provide a simple framework to analyse the growth rates of the components Xj t and their contribution to the growth rate of aggregate GDP Y t , and to identify the contribution of sector shares, the variances of the components X • t , and the co-variances of the components X. t to the volatility of aggregate real GDP.
The remainder of the paper is structured as follows. Section 4 examines the contributions of each component of expressions (4) and (6) to total real GDP growth while Section 5 examines the contributions of each component of expressions (5) and (7) to the variance of total real GDP growth. Section 6 provides a summary of the results and conclusions.

Sector contributions to real GDP growth
In this section we consider the contribution of each production sector to real GDP growth, both in their own terms and moderated by their share of total GDP. The contribution of E(e t ) to the mean growth rate of GDP is negligible (it oscillates around a mean quarterly growth rate of 0.03 percent with a mean deviation of 0.36 percent).

Sector growth rates
As a first glance, Figure 1 shows the quarterly growth rates for seasonally adjusted real GDP and the five production sectors since 1977:3. All charts in Figure 1 are drawn to a common scale and therefore provide an illustration of the relative volatility of the growth rates of each sector. For instance, the Construction sector appears to be the most volatile sector followed by the Primary sector. It is difficult however using the actual quarterly growth rates to discern changes in the pattern of volatility over time, apart from perhaps the Government and community services sector, which displays a rise in the variability of quarterly growth since the early 1990s. For this reason we use a moving average measure of quarterly growth rates, variances and co-variances throughout the remainder of the paper. An 11-quarter moving window was used throughout the paper to derive centred moving average real GDP growth rates, growth variances and co-variances. Experiments with shorter (7-quarter) and longer (I5-quarter) windows suggest that our conclusions are robust to the choice of window length.*> Centred moving average quarterly growth rates for the five aggregated production sectors and total real GDP are displayed in Figure (4) and (6).
The relative mean growth rates and the patterns of quarterly growth displayed in Figure 2 explain the changing sector shares revealed in Figure 3. They also provide some initial insights concerning the contributions of each sector to total GDP growth, the growth variance of each sector, whether the variances have changed over time, and whether sector growth rates are correlated.
The Primary and Services sectors have experienced the highest average growth rates over the last 25 years. The average quarterly growth rates were 0.73 percent for Primary and 0.65 percent for Services. Government and community services grew by 0.56 percent per quarter; Construction grew by only 0.30 percent and Manufacturing by a mere 0.29 percent. But there are marked changes in the pattern of growth rates over time. s. 19B0 1985199019952000Construction 19Í0 IMS 199019952000 1980 1915 1990 1995 2000 Construction 1S80 19BS 19901S80 19BS 19951S80 19BS 20001S80 19BS 19801S80 19BS 19851S80 19BS 19901S80 19BS 19951S80 19BS 2000 Total real GDP appears to have moved through three phases over the last 26 years. The first phase in the late 1970s and early 1980s appears to have been one of relatively volatile positive average growth; the second phase from the mid 1980s to 1990 one of steadily declining growth; the third phase commencing after the early 1990s is one of higher average and perhaps less volatile growth.
The emergence of these phases appears to be determined largely by the Services sector, the Manufacturing sector and Construction. These phases are not so apparent for the Government and community services sector. This sector appears to display two phases: below average growth in the period prior to the early 1990s; above average growth since the early 1990s, a pattern accentuated by a further rise in growth since 2000.
Primary has been the fastest growing sector during the last 26 years. Average growth and volatility appear not to have changed markedly over the period. This sector displays a more consistent cyclical pattern and doesn't always match the phases displayed by the other sectors, although the correlation of the Primary sector growth rate with the growth rates of the other sectors may have increased since the early 1990s (see Figure 4).

4.2
Sector shares These disparate growth rates generated changes in sector shares over time. Figure 3 shows how the shares for the five production sectors have changed since 1977.
It is immediately apparent from Figure 3 that the sectoral make-up of the New Zealand economy has changed over the last 26 years, especially since the mid 1980s. The Services and Primary sectors have increased their shares, the shares for the Manufacturing and Construction sectors have declined by several percentage points while the share for Government and community services has declined only slightly.
The Primary sector share of total GDP increased from around 7 percent of GDP in 1977 to around 9 percent by the early 1990s where it has remained. Services, Manufacturing and Construction sectors display sharp breaks in their shares of total GDP during the mid to late 1980s, years when major economic reforms were introduced in New Zealand. The Services share increased sharply after 1985 and has maintained a steadily increasing share in subsequent years; it has risen from around 48 percent in 1985 to nearly 53 percent by June 2003. In marked contrast, the Manufacturing share declined sharply after 1985 and its share has generally continued to decline subsequently, apart from a brief rally during the early 1990s. The share oí Manufacturing has fallen from around 17 percent in the early 1980s to 13 percent by June 2003. The share of the Construction sector fell sharply in 1989 from around 5 percent to around 4 percent where it has remained. The Government and community services share of GDP declined steadily during the late 1970s to the mid 1980s from around 17 percent to below 16 percent in the late 1980s. Since then its share has gradually increased back to nearly 17 percent by June 2003.

Sector contributions to GDP growth
This section brings together the components of total GDP growth examined in the previous two sections and described by expressions (4) and (6): the sector shares, p jt , and sector growth rates, AlogXy,. Note that the contributions measured here are the proximate contributions to total GDP growth and do not reflect the importance of a sector in terms of initiating or under-pinning growth in other sectors.
Over the full sample period, real GDP grew by an average of approximately 0.55 percent per quarter. The Services sector accounted for over half of this growth, contributing an average of 0.32 percentage points to total GDP per quarter. This reflects its substantial and rising share of GDP and the fact that it was the second fastest growing sector (after the primary sector). Government and community services was the second highest contributor with an average of 0.09 percentage points. The much smaller Primary sector was the fastest growing sector contributing 0.056 percentage points to overall growth per quarter. The larger but relatively slow growing Manufacturing sector meant it contributed less to overall growth than the Primary sector, at an average of 0.043 percentage points per quarter. The contribution of the Construction sector was close to zero.
However, there have been substantial variations about these average contributions to growth. This is illustrated by Figure 4 which shows the share weighted contributions of each sectors quarterly real output growth rate to total real GDP growth in each quarter. Figure 4 illustrates the importance of the contributions of Services and Manufacturing in determining the three phases of New Zealand GDP growth since 1977. Although Construction displays a similar pattern, its small size means that it contributes less to total GDP. The contribution of the Government and community services sector increased after the early 1990s. The contribution of the Primary sector has fluctuated throughout the period, but has become more synchronised with the Services, Manufacturing and Construction sectors since the early 1990s. The lines are: dash grey, Services; dash black, Government and community services; solid grey, Primary; solid black, Manufacturing; and dotted, Construction.
A 0.6 percentage point contribution to total GDP growth is equal to 0.006.

Sector contributions to real GDP volatility
Volatility of New Zealand GDP, measured by its 11 quarter centred moving standard deviation, has declined over the period 1977 to 2003. Figure 5 shows that the standard deviation of the quarterly growth rate for total GDP fell in the late 1980s and has remained less volatile since then. Between 1977 and the late 1980s, the average standard deviation of the quarterly growth rate for total GDP was around 1.2 percent. Since the late 1980s, the standard deviation has averaged around 0.8 percent. This represents a 33 percent fall in the average standard deviation of New Zealand's real GDP. There are several potential proximate causes of this fall in GDP volatility. GDP is an aggregate of component production sectors. Therefore its volatility is a function of the share of each sector, the variance of the growth of each sector, and the co-variance between the growth rates of each sector. This relationship is set out in expression (7) with the impact of the error term shown in expression (5). We now tiirn to examine the relative importance of these factors.

J. / Sector volatility
The behaviour of the variances of each sector is illustrated by their 11-quarter centred moving standard deviations shown in Figure 5. These plots show noticeable falls in Manufacturing and Services sector production variances during the mid and late 1980s respectively. By contrast, there is no apparent trend change in the variances of production in the Primary and Construction sectors and there was an increase in the variance of the Government and community services sector after the early 1990s.
The influence of these changes on total GDP volatility depends on the share of the sector in total GDP. Figure 3 shows that these shares have changed markedly. Each sector's moving standard deviations weighted by their share in GDP is plotted in Figure 6. The largest contributions to the change in total GDP volatility have come from the Services and Manufacturing sectors." Although the Construction sector was only around 4 percent of GDP by the end of the period, its high variance means that its influence on total volatility as measured by its weighted standard deviation is only slightly below the much larger Manufacturing sector. The Primary sector at around only 9 percent of GDP has similar and sometimes a higher weighted standard deviation than the Manufacturing sector and even the Services sector, which accounts for around 50 percent of GDP. Despite its relatively large size, since the 1980s the Services sector has not influenced overall aggregate GDP volatility any more than the other much smaller sectors.

Services sector volatility
Overall the standard deviation of the Services sector fell from around 1.2 percent prior to the mid 1980s to around 0.6 percent in the 1990s, a fall of approximately 50 percent. There is evidence in Figure 5 of a further decline since the late 1990s. Figure 7 shows a breakdown of the Service industries weighted by their share of GDP. Figure 7 illustrates that the Combined wholesale trade industry, which includes wholesale trade, retail trade and accommodation, restaurants and cafes, appears to be largely responsible for the fall in Services sector volatility since the late 1980s. The large spike around 1986 is associated with the introduction of the 10 percent Goods and Services Tax (GST) in October of that year. Seasonally adjusted real output increased by 5 percent in the September quarter immediately preceding the introduction of-GST. It fell by 6 percent in the subsequent quarter. The spike around 1989 is associated with the introduction of an additional 2.5 percent GST rate in that year. The lines are:dash grey, Services, dash black, Government and community services, solid grey, Primary, solid black, Manufacturing, and dotted, Construction.
Furthermore, there appears to have been a more gradual underlying decline in the volatility of production in the Combined wholesale trade industry during the late 1980s. The reasons for this are unclear, but this industry has experienced declining inventory to sales ratios since the mid 1980s and volatility for the trade industries may have declined as a result of improvements in inventory management techniques (see Buckle, 2000, Table 1).
The Finance and real estate and business services industry also contributed to the fall in Services sector volatility, although this appears to be due to a large spike in volatility during the early 1980s rather than a trend change. This early 1980s spike was associated with comprehensive price, wage and interest rate regulations that imposed varying degrees of limitations on price, wage and interest rate changes between June 1982 and July 1984 (Boston, 1984, Chapter 9), thereby placing greater pressure on quantities to adjust. Since the removal of those regulations, production in the Finance and real estate and business services industry has been more stable. However, as with the Combined wholesale trade industry, there is a spike associated with the introduction of GST in October 1986, and volatility in the Finance and real estate and business services industry increased again in the mid to late 1990s.
The Electricity, gas and water and Transport and storage industries have contributed relatively little to overall Services sector production volatility, except in the early to mid 1990s. The rise in volatility from the early to mid 1990s is associated with a sustained period'of economic expansion, deregulation of the sectors and, in the case of the Electricity, gas and water industry, a drought that led to low lake water storage levels and power shortages in the early 1990s (see M-co, 2001). By the end of the sample period the weighted standard deviations for both industries had subsided again.
In contrast, the Communications and Owner occupied dwellings industries have contributed very little to the changes in Services sector volatility. Owner occupied dwellings produces a stream of housing services from the housing stock. Because the housing stock does not fluctuate significantly, this stream of services is also very stable. The Communications industry growth rate has a relatively low standard deviation and has a relatively small, but rapidly rising, share of GDP. Low volatility of this industry is associated with a period of marked technical change, which may be masking cyclical effects.

Manufacturing sector volatility
The average standard deviation of the quarterly growth of Manufacturing production fell from around 2.4 percent between 1977 and the early 1980s to around 1.7 percent thereafter, a fall of approximately 30 percent. This fall in Manufacturing sector volatility is due primarily to the occurrence of two large volatility spikes for two industries during the mid to late 1980s. Figure 8 shows selected manufacturing industry weighted (by share of GDP) standard deviations. Two industries, Machinery and equipment manufacturing and Other food manufacturing experienced sharp increases in their standard deviations in the 1980s that have not appeared again.  The sharp increase in the Machinery and equipment manufacturing standard deviation occurred in the early to mid 1980s, a period associated with significant investment in large capital projects during the development of the "Think Big" projects (see Silverstone, Bollard and Lattimore, 1996). This spike appears to have been preceded by high volatility in the late 1970s. However, the short period of available data makes it difficult to determine whether the spike in the Machinery and equipment manufacturing industry is a one-off or is representative of the volatility in this sector prior to 1977.
There is also a spike in the standard deviation of Other food manufacturing. This is associated with the introduction of a 10 percent Goods and Services Tax (GST) in October 1986. There is little evidence of a decline in volatility in other manufacturing industries. The lines are: solid black, Machinery and equipment manufacture; solid grey, Printing and publishing; dash grey, Other food manufacturing; and dotted, Wood and paper products manufacturing.

Contributions of sector shares, variances and co-variances
As we have stressed, GDP volatility will be determined not only by the variances of each of its component sectors discussed in the previous section, but also by changes in the share of each sector in total GDP and the co-variance between sectors. In this section we evaluate the relative importance of each of the three components that make up expression (7). We first examine the contribution of changing sector shares. The potential for sector composition changes in the form of changing sector shares to impact on aggregate GDP volatility was emphasised by Arthur Burns in his often-cited 1960 address to the American Economic Association (Bums, 1960). If a relatively stable sector displaces a more volatile sector, everything else constant, aggregate volatility will fall (and vice versa).
The five sectors in order of volatility from highest to lowest are Construction, Primary, Manufacturing, Services, and Government and community services. From 1977 to 2003, the sum of the share changes has been to replace around 4 percentage points of Manufacturing, 1 percentage point of Construction, 2 percentage points of the unallocated and close to zero change in the share of Government and community services, with about 5 percentage points of Services and 2 percentage points of Primary. The general pattern therefore has been for a less volatile Services sector to be replacing more volatile sectors including the Construction and Manufacturing sectors. However, there has also been an increase in the share of the relatively volatile Primary sector and little change in the relatively stable Government and community services sector.
To evaluate the impact of changing shares, actual real GDP volatility (using the actual evolution of sector shares) is compared with GDP volatility simulated by weighting sector variances and co-variances using sample period mean sector shares. That is, we compare the derivation of the expression (7), the variance for S,, using the actual values for the respective sector shares, p.,, with the derivation of (7) using the sample period mean of the respective sector shares. Figure 9 illustrates the result of this comparison. It is clear that apart from some minor differences during the early 1980s, the change in sector shares has not had a significant impact on the evolution of real GDP volatility. 9 The lines are: solid, with moving shares and dotted, with constant shares.
Turning now to the contribution of sector variances and co-variances, it is apparent from Figure 5 that the behaviour of sector variances changes substantially over time and that this change varies across sectors. Hence, even with constant shares, idiosyncratic sector variance behaviour could result in offsetting influences on total real GDP volatility. Similarly, if sector growth rates are independent, the measured co-variances will be small and would contribute very little to GDP volatility. On the other hand, high dependence between sector growth rates will increase the potential for sector growth rate changes to influence total GDP volatility. Moreover, as the positive correlation between sector growth rates increases, sectors will have a greater tendency to move together leading to larger fluctuations in aggregate GDP and higher GDP volatility.
It turns out that the influence of both the sum of (weighted) sector variances and the sum of (weighted) sector co-variances are important. Figure 10 illustrates the respective contributions of sector variances, co-variances and changing sector shares to the variance of 5",, where S, is the weighted sum of sector growth rates. and excludes the error term e t which principally accounts for the difference between seasonally adjusted chained weighted GDP, Y t , and the seasonally adjusted chain weighted components.^ This diagram is constructed as follows. The plots for sector variances and co-variances are derived by summing the weighted sector variances and co-variances using constant (sample period mean) sector shares. The remaining component is the impact of changing sector shares. The lines are: black, total variance with moving shares (sum of sectors); black dashed, sum of sector variances (mean shares), grey, sum of sector co-variances (mean shares); black dotted, effect of changing shares relative to mean shares (that is, the difference between total GDP variance with moving shares and sum of sector variances and covariances with mean shares).
It is clear from Figure 10 that the decline in real GDP volatility in the late 1980s was associated with a sustained decline in the sum of sector variances and the end of a temporary spike in sector co-variances. The sustained decline in the sum of sector variances was, as discussed in Section 5.1, the result of declining Services and Manufacturing sector variances.
There appears to be no trend decline in the sum of sector co-variances series, which exhibits significant recurring cycles. These cycles in the sum of sector covariances have been the dominant influence on the profile of total real GDP volatility, especially since the decline in total volatility in the late 1980s. For example, the sharp fall in GDP volatility in the mid 1990s was primarily the result of a temporary fall in the co-variance of growth between the sectors. This suggests that at least for the last 12 to 15 years it has been the interaction between the sectors as opposed to changes in the variance of the sectors that has been the main influence on changes in total real GDP volatility.
The apparent lack of a trend and the cyclical nature of the sum of sector covariances series suggests that, barring a significant change in weighted sector variances, GDP volatility can be expected to cycle with the degree of correlation between the sectors. Furthermore, if the decline in sector variances is a permanent feature, recurrent cycles in GDP volatility may in future be around a lower trend level of real GDP volatility, provided the co-variance between sector growth rates does not increase significantly.
The decomposition of GDP volatility into sector shares, variances and covariances has revealed some unexpected results. Despite changes in New Zealand's industrial structure, changes in sector shares have not had a significant influence on the changes in GDP volatility. The main reason for the sustained decline in GDP volatility has been a decline in the sum of sector variances. There has been no significant trend decline in the influence of sector co-variances, which are the dominant influence on the profile of GDP volatility.

Summary and conclusions
Using chain-weighted estimates of production based real GDP, we find that New Zealand's real GDP has become less volatile since the mid 1980s. The average standard deviation of real GDP growth has fallen by about 33 percent since the mid 1980s compared to the average standard deviation for the period from 1977 to the mid 1980s. This decline in volatility coincides with reported reductions in GDP volatility during the mid 1980s in several other developed economies, including Australia, USA and the United Kingdom. We do not have available a sufficiently consistent long-term quarterly real GDP series prior to 1977 to enable a rigorous test of whether this decline in New Zealand's real GDP volatility is peculiar to the period since the mid 1980s or is one stage in a longer term decline that Blanchard and Simon (2001) claim occurred in several other developed economies.
To understand why New Zealand's real GDP has declined since the mid 1980s, this paper decomposes the variance of real GDP into the contribution of production sector shares, variances and co-variances. The key conclusions to emerge from this procedure are as follows. Differences in sector growth rates have resulted in marked changes in sector shares. The Primary and Services sectors have been the fastest growing sectors and the sectors with rising shares of GDP. Manufacturing and Construction have been the slowest growing sectors and these have experienced falling shares.
Despite changes in sector shares, the contribution of changing shares to changes in total GDP volatility has not been significant. The scale of the changes in sector shares is small by comparison to the variance of sector growth rates. But other factors are also apparent. The impact of the rise in the share of Services, a sector with declining volatility, has been offset by the fall in the share of Manufacturing, also a sector with declining volatility, and the rise in the share of the relatively volatile Primary sector.
The decline in real GDP volatility was due primarily to a sustained decline in the sum of sector variances. Two sectors were key contributors to the decline in sector variances: Services (specifically the Finance and rea! estate and Wholesale trade industries) and Manufacturing (specifically the Machinery and equipment manufacturing and Other food manufacturing industries). There was no obvious change in the behaviour of the relatively volatile Primary and Construction sectors. In contrast, the volatility of Government and community services increased after the early 1990s.
Throughout the sample period, changes in sector co-variances have been the main factor determining the profile of aggregate real GDP volatility. This influence has not diminished and its relative significance has increased.
The industry evidence presented in this paper suggests that industry, regulatory and fiscal policy interventions during the early and mid 1980s explain most of the relatively high volatility during that period compared to subsequent years. Policy interventions that appear to have been most important in contributing to the relative high volatility of the 1980s include the "Think Big" industrial development strategy, a comprehensive price, wage and interest rate freeze introduced in 1982, and the introduction of the Goods and Services Tax (GST) in 1986 and 1989.
Cessation of these policy interventions, responses to deregulation of many parts of the Finance and real estate industry, and the impact of more widespread deregulation on the behaviour of the Wholesale trade industry and industries in the Manufacturing sector appear to be key reasons for the decline in real GDP volatility since the mid 1980s. Changes in inventory management techniques may have also played an important role in some industries, particularly the Combined wholesale trade industry. To the extent that these regulatory and industry changes are permanent features, the New Zealand economy may indeed have entered a period of calm following the relatively stormy years prior to the mid 1980s.
These conclusions contrast with those of some studies that have examined the reasons for the decline in GDP volatility in other developed economies and which have identified reductions in supply shocks and.the conduct of monetary policy as key contributors to changes in volatility. Our decomposition methodology may not be the ideal approach for identifying the precise contribution of supply shocks and monetary policy, but we think it nevertheless provides some useful insights concerning their relative importance.
For example, the timing and industry specific nature of the decline in sector variances suggests monetary policy and its focus on inflation targeting may be the predominant explanation for the initial decline in New Zealand's GDP volatility. But if monetary policy was the predominant influence, we might expect to see that influence reflected in a change in volatility across several sectors, and in particular in sectors that would be expected to be more sensitive to interest rates, such as the Construction sector. Furthermore, although formal inflation targeting has been associated with a fall in the variance of real GDP, the timing of the decline in real GDP volatility predates formal inflation targeting. * * For similar reasons, it would be difficult to infer from our procedure the full impact of fiscal policy on GDP volatility. Some of the explanations for lower volatility are associated with cessation of changes to fiscal policy instruments. On the other hand, the volatility of the Government and community services sector increased after the early 1990s. Nevertheless, the contemporary focus of fiscal policy on issues of sustainability and inter-temporal equity may have enhanced the tendency toward less abrupt changes to other aspects of fiscal policy. This shift in approach to policy may in fact have been more important than the indirect influences on GDP volatility typically considered arising from the impact of a longer-term focused fiscal strategy on private sector behaviour (see for example Allsopp and Vines, 2000).
This potential effect appears to have been largely ignored in the contemporary debate about the reasons for the decline in GDP volatility across so many countries. Furthermore, the switch to less frequent and less abrupt policy interventions that has occurred in many developed countries might provide an important clue to the close timing of the decline in GDP volatility across several developed economies during the 1980s.
Our.decomposition of GDP volatility raises other interesting questions that warrant further investigation. Why have the significance of the co-variances held up and their relative significance increased? What drives changes in the sector covariances? Uncovering the factors that determine the variability of sector covariances warrants deeper investigation because it is the most important component determining the cyclical variation in GDP volatility. We postulate that shocks that are common to many sectors, such as changes in interest rates, the exchange rate and aggregate demand will tend to raise the sector co-variance and raise GDP volatility. Shocks that are sector specific, such as sector specific climatic shocks and productivity changes, could be expected to have a smaller impact on aggregate GDP volatility.
Has deregulation changed the way sectors interact and influence GDP volatility? Preliminary cross-correlation analysis reveals a fall in the correlation between Services and Manufacturing sector growth rates and a rise in the correlation between the Primary and Services sectors since the early 1980s, suggesting either a change in the linkages between these sectors or that the type of shocks that have impacted on these sectors has changed. These are some of the issues that remain challenges for future research.