Keywords: Growth & Innovation; Statistics;
Sources of Economic Growth in New Zealand: A Comparative Analysis is a paper attached to the latest IMF review of the New Zealand economy, prepared by Abdelhak Senhadji who was one of the IMF review team. It is on the IMF website After reviewing the record of New Zealand’s slow growth performance it presents a (reduced form) econometric equation which attempts to provide quantitative estimates of various influences on New Zealand’s poor performance. This paper reviews the study.
Notes: 1 . The paper is in a preliminary form and will be prepared for publication. Its writer saw an earlier version of this paper, and will it into consideration for revision. He reminded me, though, that sometimes I offer an impractical counsel of perfection.
2. The paper’s disclaimer that ‘it is important to stress that the results need to be interpreted with caution given that the relationship between the explanatory variable and GDP per capita growth in the estimated equations is not necessarily causal.’ While keeping the caveat in mind, I propose to proceed as though the causality is there.)
The Data Set
The data set is derived from a panel of 21 OECD countries for the period from 1971 to 2002, some 630 observations in all, the individual data items mainly coming from an OECD data base. How adequate is the data set?
But to begin with a methodological problem, which is not so much about the equation as about its use. The equation may find regularities across the panel of countries, but that does not mean it is necessarily useful for examining one or a subset of the data points in the panel. In particular the application to a single country may be inappropriate. My comments, which are about the relevance of the results to New Zealand, must be seen in this context.
The OECD Data Set
This data set is one of the best quality in the world, and I propose not to quarrel with it. It should, however be noted that very often National Accounting data for the December (calendar) year T is in fact for the March year T+1 in the case of New Zealand.
The Choice of Time Period
The time period beginning in 1971 reflects what is available for the OECD, but is unfortunate for New Zealand, because it experienced a major terms of trade structural shock in 1966 which the economy spent the next decade and more adjusting to. As I have said in another context, examining the New Zealand economy from 1971 is like studying the economy of Hiroshima from 1950.
Because of data base limitations, we cannot do much about the beginning date. But we need to be cautious about applying the resulting equation to New Zealand.
The Country Panel
The more I look at the OECD the more I am struck that there are two sorts of economic growth. While most grow at roughly the same ‘cruise’ rate as the rest, there are a handful which experience ‘turbo’ growth. Turbo growth is characterised by very rapid growth from a low level of per capita GDP, often doubling GDP in a decade. However, once the level of per capita output reaches that of the top countries, economic growth rates level at the cruise rate and so the economy joins the rest of the OECD. Examples include Germany in the 1950s, postwar Japan to 1990, Ireland and Spain in the 1990s, many East Asian economies including Korea in the 1980s and 1990s. Cruise growth is when the economy joins the top OECD bunch measured by GDP per capita (and where there maybe considerable doubts of exact rankings because of measurement problems). At this point the growth rates are similar and presumably so are the determinants of the growth rates.
Every country wants to instigate turbo growth, but the evidence is that once they join the throng at the top of the rich OECD this does not happen
The econometrics usually allows for this by the convergence effect (discussed below) but in my view that is inadequate. Turbo growth is fundamentally different from cruise growth. It is instructive that once the economy has joined the cruisers, the policies which seemed so effective under turbo conditions become as ineffective as they are for other cruisers. Turbo growth needs to be studies separately. My guess is that is usually depends upon the rapid absorption of underutilised resources (including underutilised skilled employed labour as well as unemployed labour) and the import of quality technologies (and often foreign capital).
The “New Zealand problem’ may be posited as to why New Zealand has not maintained a cruise growth rate in much of the post-war era, but has tended to be below it on average. My research shows that New Zealand has usually cruised at a similar rate to the rest, except on three or four occasions it stagnated for long enough to cause its relativity to other cruisers to change. I argue that these occassions are associated with a major shock to the external sector, either a fall in the terms of trade or a rise in the real exchange rate (both of which are equivalent to a fall in the profitability of the tradeable sector.) Whatever, the issue which arises out of the New Zealand problem is to compare New Zealand with other cruisers, not with the turbos. New Zealand’s turbo growth experience was from 1935 to 1945. The issue of turbos versus cruisers is a separate research project which does not particularly involve New Zealand. The New Zealand problem is why it was a slow cruiser.
Consequently, my preference would be to exclude from the panel countries Ireland, Japan, Korea and Spain (and possibly Norway because of the impact of oil finds on the economy). The easy prediction is that the ‘convergence effect’ will no longer be econometrically significant. The interesting question is whether any of the other coefficients will move.
Equation Specification
The equation consists of an independent variable change in (log) per capita GDP , which measures the GDP per person growth rat, with more than 13 independent variables discussed below.
Note that when the independent variable is a also change variable, say the change in (log) X, then the relationship can be integrated back and discussed as the level in (log) per capita GDP affected by the level of (log) X, so the regression coefficient is the elasticity between the variables (before logging). This is the form that I investigated. However the data in my equation was co-integrated order one, so I in effect estimated a similar ‘change’ equation.
Change in (log) per capita GDP: The difficulty with this variable is that the theory would suggest that rather than per person, it would be more appropriate to divide GDP by per worker, or better still, per hour worked, so that labour productivity is being measured (that being the relevant variable from the production function). The first is available in the OECD data base, but the available hours worked series are not reliable (and may not cover the whole period).
Additionally, the theory – confirmed by Steve Dowrick econometrically – says that (relative) population growth has an impact on GDP . Omitting this variable may cause population bias.
Now follows the independent variables.
(Log) GDP per capita in 1971: This is a levels variable. Its function is to test the convergence hypothesis, that is that poor economies grow faster until they catch up to the rich economies. I suggested earlier that this is a proxy for turbo performance, and it would be better to deal with that phenomenon separately.
Change in (log) employment: as mentioned, it may be theoretically more satisfactory to include this as a part of the dependent variable.
Change in (log) quantity of capital: while this is a theoretically correct variable to include (although some would think it should be included in a dependent variable which would better be measuring total factor productivity). However I have some concerns about its measurement.
To give a simple anecdote, when the price of oil rose in 1974 Japan quickly closed down most of its oil-fired smelters, which while not technologically obsolete were priced out of the market. However in the normal way capital is measured those smelters would have remained in the estimates of capital stock until they were slowly depreciated out. In the case of Japan such events may have be so small relative to the total as not to markedly effect aggregate capital stock, but in the case of New Zealand various events in the period were sufficient to affect the total. These include
– the dramatic collapse in the price of wool in the late 1960s which made parts of the capital invested in sheep farming obsolete;
– the removal of border protection in the 1980s and 1990s, which made parts of manufacturing capital obsolete;
– the fall in the price of oil in the mid 1980s which made obsolete parts of the gas-based petrochemical industry developed in the early 1980s.
Probably most of these early obsolesences had worked their way out by 2002, but the profile of available capital stock over the period does not the actual New Zealand measure.
It should be noted that the sum of the estimated coefficient on capital and the coefficient on employment come to between .32 and .35 in the four equations. That means that a 1 percent rise in employment and capital would raise output by only .35 percent at most.
Inflation: The inflation variables (explained below) are measured by the CPI. I am never sure the of relevance of this measure, since the CPI is not a measure of production inflation, it being greatly affected by import prices. (The path of the New Zealand CPI shows considerable divergence from that of GDEF, the GDP deflator.) Since a GDEF is available in the OECD data base,, and it is the relevant production side variable, why not use that?
The equation uses two inflation variables, because it believes there is a curvilinear relationship between economic growth and inflation. While I am not antagonistic to that hypothesis, it is perhaps unfortunate the way it was tested (even aside from the wrong price measure). It assumed the turning point was an inflation rate of 3 percent p.a. and then had a linear function on both sides, instead of putting in a quadratic function. For the record, the regression estimates suggest that increase of inflation from 0 to 3 percent p.a. would add .6 to .8 percent to annual GDP growth.
However because of the asymmetry, inflation at 10 percent p.a. would give as much growth as inflation at 0 percent p.a. I am not sure I believe that, but even so it suggests that the 1 to 3 percent inflation target is wrong and a better one might be 2 to 6 percent p.a., with a ‘midpoint’ of 3 percent p.a.
I worry whether it makes sense to combine the high inflation era of the 1970s and early 1980s with the low inflation era which followed. It is possible that a dummy variable splitting the period into two would do as well or better. In some respects for a small country may be concerned more with the relative inflation rate to other OECD economies.
Savings Ratio: I have not spent enough time on the OECD measure of the national savings ratio to comment on this independent variable. Since it includes public savings, and assuming any Ricardo effect is negligible, an implication is that the New Zealand government running a 7 percent of GDP budget surplus seems to be adding about .2 percent p.a. growth to GDP. (I will not comment on the implication of the US fiscal position.)
Change in (log) Relative Unit Labour Costs: Again I am not sure how robust this measure is (since if we dont have a good track on hours worked, how can we have one on the hourly wage?). The regression estimate suggests that a ten percent annual wage hike reduces GDP growth by about .5 percentage points, a figure which many unionists would find attractive.
Change in (log) Terms of Trade: This variable is theoretically justified and significant. However its effect seems very small, for it says that a 10 percent reduction in the terms of trade will reduce the level of GDP by about .2 percent. In the case of New Zealand that is almost certainly too small. (I have independently estimated the impact is closer to 6.2 percent).
The trouble may be that there is no lag structure in the model, for a terms of trade shock may take effect over a number of years (and indeed initially be perverse in the case of livestock slaughter). It is also possible that the effect is non-linear. In particular the New Zealand economy has probably accommodated to the steady but annually small decline in the terms of trade for meat and some dairy products. It is the big shocks which have upset it.
The estimation included ‘two dummy variables for the oil shocks in the 1970s’ but not one for the reverse oil shock of 1985. The estimates are not given so the magnitude cannot be commented upon. They probably need to be thought of as system shocks rather than terms of trade ones, since oil price changes already appear in the terms of trade. (This might be the reason for omitting the 1985 shock.) In some ways an oil price shock to the world is analogous to the wool price shock to New Zealand.
If the terms of trade are a relevant variable, then so too is the real exchange rate, since there are different aspects of the profitability of the export sector. (They are different because the real exchange rate includes the price of non-tradeables, which is not in the terms of trade, and because it treats export and import prices together rather than in ratio.) I have used for the real exchange rate the ratio of GDEF (GDP deflator) divided by the harmonic average of the price of exports and imports, which is available from the OECD data base. In the New Zealand case the effect of the rising real exchange rate seems important (indeed to my astonishment, more important than the terms of trade in the long run). (See External Impacts and the NZ Economic Growth Rate. ) It will be returned to below, both in the discussion of the New Zealand dummy and in the discussion on mis-specification.
It is also appropriate to put on record an issue which has begun to worry me (and others). The OECD terms of trade applies to merchandise exports and imports only. How are we to allow for the rising proportion of services in international trade?
Share of Government Consumption in GDP: The trouble with this variable is that even though it omits transfers (which seems wise given the complicated but differing ways that they impact in different countries), different countries have different conventions in the measurement of government consumption. In particular, New Zealand levies its GST (a Value Added Tax) on all government consumption in exactly the same rate as it levies it on private consumption. This may have the effect of artificially raising the New Zealand ratio by 12.5 percent or about 2.5 percentage points of GDP relative to many other jurisdictions. However the effect of the ratio seems small. (It is unclear what the units of measurement are, but if they are a pure ratio, a country with a 1 percent of GDP higher spending ratio would have its growth rate reduced by .00002 percent p.a., if it a percent – as hinted in Table 2 – it would be .02 percent, which makes some contribution to explaining the dummy and distance effects described below).
Tax Variables: Two tax variables are included – the highest marginal individual and corporate tax rates. Again depending upon the way each variable is measured, their effect may be very small.
Openness: This variable is measured by the ration of exports to GDP. While this does not directly affect New Zealand, a number of countries are entrepot centres – importing and exporting the same products. That seems to me to be quite a different activity to what happens in NZ. Either the variable should be adjusted for re-exporting, or perhaps treated as non-linear.
There is also a deeper problem here which I have not seen addressed. On this measure America would appear to be a relatively closed economy. But each of the states of the US would be relatively open. On the other hand, individual states of the European Union are treated as separate observations, which gives the impression that Europe is more open that the US. How to deal with this aggregation problem?
(The point here is not to challenge the notion of openness leading to growth, although the effect may be curvilinear, in the sense that a more open economy is likely to grow faster – at least if it is small. The issue is how to measure openness.)
I am surprised that openness has been seen to lower New Zealand’s GDP growth rate. By the mid 1990s the export to GDP ratio was about 32 percent, above the (weighted) OECD average of 19.5 percent (and the US of 12.5 percent).
Other variables: the oil shock variables have already been mentioned. There is also a time trend and an adjustment for the business cycle.
There is no indication whether the time trend is econometrically significant. Presumably it refers to a belief that there is a secular slowing in the average cruise growth rate of all the rich OECD. However some would argue that while that may have happened in the 1970s and 1980s there was some growth acceleration in the 1990s.
The report mentions that ‘all variables which are business cycle sensitive have been pre-filtered with the HP filter to remove business cycle frequencies.’ Unfortunately in New Zealand there is no single business cycle frequency., the length of the cycle varying between two and a bit and almost seven years.
The New Zealand Dummy: In two of the four equations a dummy variable for New Zealand is added. It suggests that New Zealand tends to grow slower than the average OECD, after adjusting for all these other effects, by about .5 per cent p.a. However a similar dummy is not included for any other country. This leads to the next variable, but as explained below, we are on dangerous methodological grounds.
Distance: Table 4 it suggests that Australia and the US are distant but Canada is not. I suspect there should be an adjustment for tho countries which are themselves ‘economic centres’.
There are two methodological concerns with this distance variable. The first is that it is not at all clear how distance affects the growth of GDP. It is easy enough to give an explanation, as to why the costs of distance may lower the level of GDP, but the connection to growth is much less obvious and needs to be explained.
Second, the way that the distance variable is introduced is uncomfortable. The econometrics found an anomaly of New Zealand’s behaviour, via the use of a dummy variable (but it did not check whether any other country was also anomalous). It then, in effect replaced the dummy variable by a pseudo-dummy which has many of the characteristics of a dummy variable since it is constant throughout the period impacts, largely on a single country New Zealand (but also Australia), and which seems mis-specified. However no other possible variables are tested – I would predict that the proportion of the population who play rugby is likely to be about as equally effective. But having identified one pseudo-dummy it is easy enough to identify others. (It would also be appropriate to redefine the varibale so that large economies are defined as close to themselves.)
In practice the difference could be due to measurement difficulties (recall the problem of the measurement of government spending) or an omitted variable. In the latter case a candidate must be the real exchange rate, because it is difficult to explain the course of the New Zealand economy from the late 1980s without it.
What seems to be happening here is that there may be an omitted variable, but the distance variable is far from proven. We turn now to specification problems.
Specification Problems
Omitted variables
The theory of econometric mis-specification concludes that omitting a relevant variable is likely to bias the remaining parameter estimates, while including an irrelevant variable will not. Omission, then, can be a very serious problem.
Among the potentially important variables that appear to be omitted are:
1. Relative population growth;
2 The real exchange rate;
and possibly a
3. A business cycle or capacity utilisation variable (although there may be no suitable candidate in the OECD data base and the filtering procedure used may be the best available, if markedly inferior).
Mis-specified variables
Among the improvements that might be made to the variable specifications are:
4. Scale GDP by employment or hours worked instead of population.
5. Use GDEF (the GDP deflator) as the measure of inflation, and put it in as a quadratic form so that the ‘bottom’ is not predetermined.
6. It may be appropriate to include lagged variables.
Additionally there are other ideal improvements, but there are no suitable data series.
Observations
As already mentioned the turbo economies should be dropped from the panel (or at the very least other regression runs should be done without them).
And always the counsel of perfection is to obtain longer data series, if that is possible.
Estimation
The estimation procedure seems to be ordinary least squares. It covers variables which are co-integrated to degree zero and one. The likely effect of this mixture may be that the standard errors of the regression coefficients are biassed down, so they seem more precise (and significant) than they are.
I suspect this is so because so many of the parameters are tiny, but significant.
Conclusion
This is written from the perspective that the original report represents progress, and that by subjecting it to rigorous review, with positive suggestions for improvement, further progress can be made.
Even so, the methodological question of the usefulness of the equation to understanding New Zealand remains. In particular it seems to me that it sheds little light on the following questions:
1. Why did the New Zealand growth rate slow down at the end of the 1980s so that for five successive years there was a reduction in the GDP per capita?
2. Why was there a recovery of the growth rate to more normal levels in the mid 1990s.