# Comparative properties of models of the UK economy.

1. introduction

The nature of the association between inflation and the level of unemployment has been a persistent issue of controversy over the last three decades. initially, attention focussed on the statistical relationship between nominal wage inflation and unemployment-the Phillips curve-which could be seen equally as a relationship between price inflation and unemployment, if prices are a constant mark-up on wages. This was quickly adopted as a menu for policy choice, describing the trade-off between increases in unemployment and reductions in inflation. By the 1970s, however, the question was whether a long-run trade-off existed at all, the OECD economies having experienced rising unemployment and, simultaneously, rising inflation. The subsequent re-examination of labour market behaviour introduced the concept of an equilibrium rate (the natural rate) of unemployment which, in the monetarist view, was not amenable to demand management policies. More recent developments reflect a growing concern with the supply side of the economy, including the question of what determines the non accelerating inflation rate of unemployment (NAIRU).

The emerging framework uses bargaining models for the determination of wages: unionised employees set pay demands and employers respond according to their ability to meet these demands, depending in turn on their pricing decisions. In an influential analysis, Layard and Nickell (1985) develop a three equation model, determining wages, prices and employment, and emphasise the role of 'push' variables, which can affect firms' price mark-up on wage costs or wage bargainers' mark-up on prices, that is, the real wage. The push variables include taxes and benefits, import prices, and productivity. For given values of such variables there is a unique level of unemployment, the NAIRU, which leads bargainers to settle for a real wage which is consistent with that which firms are willing to accept in their pricing behaviour. Large-scale macroeconometric models differ substantially in size from the Layard-Nickell model, nevertheless this provides a useful starting point in studying their properties, given that they increasingly incorporate aggregate supply considerations. Extensions to the Layard-Nickell framework are necessary, however, most notably to take account of the endogeneity of the real exchange rate, which is an important adjustment mechanism in an open economy with floating exchange rates. Additional push factors, such as a deterioration in the current balance, can then act to produce higher relative prices of imports, in turn increasing the NAIRU. Recent NIESR research (Joyce and Wren-Lewis, 1989; Wren-Lewis, 1989) shows how the change in the NAIRU following any policy change on the NIESR model can be derived as a simple function of the long-run solutions to the model's wage and price equations, together with the response of the real exchange rate.

This paper describes six major macroeconometric models of the UK economy, using their wage-price-exchange-rate interaction as a core supply-side framework in which to interpret their properties, as revealed in standard simulation experiments. The wage, price, and exchange-rate equations provide considerable insight into the simulation results, and these are accordingly analysed next, in Section 2. Some general features of the simulations are noted in the course of this discussion, with detailed analysis presented in Section 3. The numerical results of the policy simulations for the major macroeconomic indicators-ready reckoners-are presented in Appendix tables A1-A5, and Section 4 contains concluding comments. The models considered are those of the London Business School (LBS), National institute of Economic and Social Research (NIESR), Her Majesty's Treasury (HMT), Bank of England (BE), Oxford Economic Forecasting (OEF) and the Liverpool University Research Group in Macroeconomics (LPL), as deposited with the Bureau in late 1989; the HMT model was also publicly released at this time. All of these are quarterly models, except the LPL model, which is annual, and as this has remained largely unchanged over the last six years there is little discussion of it in the present paper: reference may be made to our earlier analyses in this Review and elsewhere (for example, Fisher et aL, 1989). The OEF model is a newcomer to our comparative analysis. 2. Wages, prices and the exchange rate Differences in the way in which wages, prices and the exchange rate are determined in the various models provide a major part of the explanation of differences in their overall properties. This section therefore presents an analysis of the relevant equations, as a prerequisite to an understanding of the policy responses.

Wage equations

The wage equation in most macro models is an important link in the resolution of the overall association between changes in unemployment and inflation. Even at the single-equation level there is wide diversity across the models in the long-run response of wages to a change in unemployment, as shown in table 1. The functional form in which unemployment enters the wage equation differs across the models, so for comparative purposes we examine the sensitivity of wages to a fall in unemployment of 100 thousand from an initial level of 2 million, allowing where relevant for interaction between wages in different sectors (LBS and BE) or for changes in the duration structure of unemployment (LBS; BE and NIESR have now removed this distinction in the determination of wages). The LBS model is a clear outlier, with the rise in total wages nearly three times greater than that in any other model. This is in turn reflected in the full model simulation results, as the rise in real wages is much greater in all the LBS simulations, by the last year. The sector which is most responsible for these effects is manufacturing, whose wages then determine those in non-manufacturing. The LBS manufacturing wage equations are very different not only from comparable equations in the other models but also from wage equations in previous vintages of the LBS model, and econometric analysis reveals severe problems; of particular importance is the instability of the estimate of the long-run real wage response to a change in aggregate unemployment. The BE wage equations by contrast have a particularly low response to changes in unemployment, but this key response is ill-determined statistically.

The functional form in which unemployment appears in the wage equation is one of the major sources of non-linearity in the models, and must be borne in mind before scaling up the ready reckoner results to evaluate much larger policy changes. in the LBS, HMT, OEF and LPL wage equations the logarithm of the unemployment rate is used, implying that the same absolute fall of 1 00 thousand from an initial level of unemployment of one million would double all the wage effects in table 1. in the BE wage equations the reciprocal of the unemployment rate is used implying, approximately, a quadrupling of the wage responses in table 1 if the initial level of unemployment is reduced to 1 million. The NIESR model is an interesting exception since, rather than the official claimant measure of unemployment, the proportion of the adult population not in work is used in the wage equation, on the grounds that this is a superior measure of wage pressure, and as this variable is not transformed the NIESR response in table 1 is not base dependent.

The single-equation response of nominal wages to changes in consumer/retail prices is very rapid across all of the quarterly models, with nominal wages rising more or less in line with prices. An important exception is the OEF wage equation, where the elasticity of nominal wages with respect to prices takes a value less than one, namely the unrestricted point estimate of 0.93.

All of the wage equations include a time trend or some measure of productivity to explain the growth in real wages over time. Of particular importance is the use of a direct measure of productivity in the NIESR model: as increases in employment tend to lag behind increases in output a sharp rise in productivity results, contributing to pronounced increases in real wages in the first two years of all the NIESR simulations, although real wages subsequently fall back as the employment response comes through. A further contribution to these short-term surges in real wages comes from a 'mismatch' variable which has a strong influence whenever employment is changing rapidly. Such sharp increases in real wages do not feature in the other model simulations since those models which include productivity in the wage equation (HMT, BE and OEF) use a trend or smoothed measure of productivity, and in no other model does a mismatch' variable influence wages. The response of real wages helps to explain (in conjunction with the exchange-rate response) why the initial effects on both inflation and demand tend to be larger in the NIESR simulations than for other models.

Price equations

In all of the quarterly models, prices are determined largely as a mark-up on costs, with a 1 per cent increase in all costs (mainly labour and import costs) eventually leading to a 1 per cent increase in prices. However for a number of the equations in the OEF price sector this condition of static homogeneity has not been imposed, even though in most cases the freely estimated coefficients do not depart substantially from this restriction. This feature of the OEF price equations together with the absence of static homogeneity in the wage and exchange-rate equations seriously dampens the transmission of inflationary pressures in this model. it is a consequence of a modelling strategy in which empirical validity has received higher priority than theoretical consistency.

The equation for manufacturing wholesale prices can be regarded as the pivotal equation in all cases, although the extent of disaggregation of the price system differs significantly across the quarterly models, as does the speed of adjustment. According to the NIESR, BE and HMT wholesale price equations, 50-60 per cent of a step increase in total costs is passed on in higher prices within two quarters, and at least 90 per cent within four quarters. The response of the OEF equation lags approximately one quarter behind these models, probably because no contemporaneous effect of costs on wholesale prices is allowed. However the main outlier is the LBS equation, since it takes six quarters before 50 per cent of an increase in total costs is passed on in higher prices and more than four years before this proportion reaches 90 per cent. This sluggishness is likely to dampen substantially any inflationary effects in full model simulations. Our re-estimation of the LBS equation suggests that these responses are sensitive to the sample period and that if this is confined to the more recent data then the responses are very similar to those from the other models.

In all of the manufacturing wholesale price equations the size of the mark-up on costs is sensitive to the pressure of demand through some measure of capacity utilisation. However from the point of view of full model simulations the way in which capacity utilisation is determined is perhaps of greater importance than the single-equation effect of any given change in capacity utilisation on prices. Thus, in the BE and HMT models capacity utilisation is basically determined as a distributed lag on manufacturing output so that any increase in output, once it stabilises at a constant difference from base, eventually leaves capacity utilisation unchanged. On the other hand, in the LBS, NIESR and OEF models, changes in capacity utilisation emerge from the dynamics of output and capital stock adjustment, as determined elsewhere in the model. In the case of NIESR this adjustment process is very prolonged and usually cyclical but does seem to ensure that capacity changes in line with output in the long run. However this does not appear to be the case for either the LBS or OEF models, where expansionary policies are associated with persistently higher capacity utilisation, which might seem to be a doubtful long-run property. By itself this would also result in inflationary effects that are stronger than in other models; that these are not observed is principally due to features discussed above, namely the very sluggish price dynamics in the LBS model, and the absence of static homogeneity throughout the OEF wage-price sector and in its exchange-rate equation, discussed next.

The exchange rate

In explaining the role of the exchange rate it is useful to assess the models' exchange-rate equations according to three characteristics: the extent to which they are forward-looking; their responsiveness to changes in competitiveness; and their responsiveness to current account imbalance.

The exchange rate in the LBS and NIESR models is forward-looking and hence in response to a shock the nominal exchange rate immediately 'jumps' part of the way (LBS), or almost all of the way (NIESR) to the long-run change in the exchange rate. The large fall in the exchange rate on the introduction of expansionary policies means that inflationary pressure from higher import prices and the stimulus to GDP via improved competitiveness and hence net exports are greater in the early years of the LBS and NIESR simulations in comparison with other models. The LBS exchange rate is determined as a market-clearing price in the financial sector of the model and hence it is not possible to analyse the precise mechanism leading to a change in this variable. Given the key role of the exchange rate in determining both short- and long-run model responses, this is a serious handicap to understanding the simulation properties of the LBS model, particularly given the criticism of many parameter values in its financial sector made by Courakis (1988). The long-run change in the NIESR exchange rate is determined so as to correct any current account imbalance, or more precisely any imbalance in net overseas assets: an ex ante deterioration in the current account balance equivalent to 1/2 per cent of GDP (about 22.2 billion in 1990 prices) leads to a step fall in the nominal exchange rate of roughly 6 per cent. Fisher and Turner (1990) show that the magnitude of this response is sensitive to those parts of the model which deal with the effect of changes in wealth, including revaluations; unfortunately these effects are among the most difficult to quantify.

The exchange rate in the BE model is specified in real terms so that in simulation the nominal exchange rate falls ceteris paribus in line with increases in (wholesale) domestic prices. This homogeneity property would appear to be one which HMT and OEF (which has an equation closely based on the HMT specification) have attempted to incorporate in their exchange-rate equations. Thus according to Melliss (1988, p.248) the HMT equation

'uses the idea that relative money supplies and relative

labour costs are an indicator of the future movements in

UK relative to world prices. The unit coefficient on the

sum of these terms in a log specification indicates that

the equation is for the equilibrium real exchange rate.'

[emphasis added]

However in the various simulations on the HMT and OEF models, relative money supplies are not associated one-to-one with future movements in prices, at least within the time period considered. Moreover the actual association varies between simulations, so that the real exchange rate may rise or fall depending on the nature of the shock. In the case of OEF the growth in broad money tends to exceed the change in prices to the extent that output rises, and so this typically contributes to a fall in the real exchange rate. However this effect is partially offset by the inclusion in the OEF exchange-rate equation of UK domestic labour costs specified in dollar rather than sterling terms. This means that following an increase in UK labour costs the exchange rate does not depreciate as it ought to do in order to maintain the homogeneity property.

The current account balance enters into the OEF and BE exchange-rate equations: following an ex ante 1 billion pounds (1990 prices) deterioration in the current balance there is a long-run fall in the exchange rate of about 0.6 per cent for OEF and about 0.2 per cent for BE. A measure related to the current balance is also included in the HMT exchange-rate equation, but its coefficient is so small that it has a negligible influence. in addition a (cumulative) measure of the PSBR is included in the BE equation, as a measure of a risk premium, so that a El billion per annum increase in the PSBR leads to a fall in the exchange rate of about 0.3 per cent after five years.

Our analysis of the wage, price and exchange-rate equations in the quarterly models has revealed a number of statistical deficiencies and theoretical inconsistencies, which except in the case of the NIESR model preclude a straight forward derivation of the models' overall long-run properties. Nevertheless an understanding of the wage-price-exchangerate interaction, including any shortcomings, provides useful guidance in comparing the results of standard simulations, to which we now turn.

3. Standard simulations

The policy simulations considered in this section comprise an increase in government current expenditure of E2 billion per annum (1990 prices) reductions in the rate of income tax, VAT and employers' national insurance contributions of one percentage point, and a cut in nominal interest rates, also of one percentage point. It is assumed that any increase in the PSBR is accommodated under constant nominal interest rates, approximating money finance, and that all changes are assumed to be unanticipated and permanent, unless otherwise stated. General results on the major macroeconomic indicators over a five-year horizon, the longest period common to all models, are presented in tables A1-A5.

Our discussion extends that of the previous section by considering, in particular, the links between inflation and demand, these being necessary if convergence to a NAIRU is to occur. In several cases we observe that there is no such tendency and we are then able to use the foregoing discussion of the wage, price and exchange-rate equations to describe the nature of the inflationary processes themselves.

Government expenditure

In this section we examine the effects of an increase in government current expenditure, where the increase is allocated proportionately between procurement and employment in those models that make this distinction (LBS, HMT and OEF) and between central and local government where this additional disaggregation occurs (HMT). The size of the government expenditure increase is El,440 million per annum in 1985 prices, a figure which has been chosen so as to produce an increase in 1990 prices of about 22 billion per annum.

The GDP multipliers (defined as the ratio of the absolute increase in GDP to the increase in government expenditure) are shown in table 2. For all the quarterly models, except NIESR, the impact multiplier in the first year is about unity. After five years the multiplier has risen to between 1.6 and 1.8 on the LBS, BE and OEF models, although much of the rise in the LBS multiplier takes place in the second year rather than gradually, as in the case of the other two models, whereas the HMT multiplier remains virtually at unity over the entire simulation horizon. The NIESR multiplier clearly shows the greatest volatility with an impact multiplier of 1.8 rising to 2.2 in the second year before failing back sharply to only 0.4 by year 5.

The increase in government expenditure on the NIESR model leads to a decline in the model's NAIRU and hence a long-run increase in GDP, because government expenditure is generally less import-intensive than private sector expenditure and so brings about an increase in the long-run real exchange rate. This in turn reduces inflation by making imports relatively cheaper and so enables a longrun fall in unemployment of about 30-40 thousand to take place without increasing overall inflationary pressure. However in the short run there is a much larger multiplier effect which follows from the effect on demand of the initial improvement in competitiveness due to the step fall in the nominal exchange rate, as well as the temporary boost to real wages, discussed above. The adjustment from the large initial multiplier to the smaller long-run multiplier which is consistent with the change in the NAIRU, takes place through the effect which higher inflation has in eroding competitiveness and crowding-out consumers' expenditure, and hence reducing demand.

Explaining the simulation results in terms of the change in the NAIRU is more difficult for the other models. Partly this is because, as previously shown in Section 2, there are a number of inconsistencies in their supply-side structure which prevent calculation of the change in the NAIRU. However a second reason is that although the NAIRU defines an equilibrium in terms of inflation and unemployment there is no guarantee that the simulation will converge on this equilibrium. The NIESR model is an outlier in this respect: as noted above and in our previous review (Fisher et al., 1989, p.73), this model rapidly converges on the NAIRU. The absence of such strong adjustment mechanisms means that for the other quarterly models it is more appropriate to explain the response of unemployment and GDP in terms of changes in the components of demand. Our supplyside analysis in Section 2 is then useful in understanding the inflationary consequences of this change in demand.

The increases in the multiplier over the course of the BE and LBS simulations largely originate from the effect on demand of the fall in the real exchange rate, as is demonstrated by the fact that variant simulations in which the real exchange rate is held at its base values deliver a multiplier response which remains close to unity. The resulting improvements in competitiveness help to restrain imports and stimulate exports, which in turn boosts both stock-building and investment via accelerator type mechanisms. These accelerator effects are particularly strong in the BE simulation, with the absolute change in fixed investment nearly as large as that for consumption. This result is explained by the equations for manufacturing and non-manufacturing investment, which have elasticities with respect to sectoral output of 1.5 and 2.1 respectively, although the corresponding elasticities in the other models are typically constrained to take a value of unity.

The improvement in the OEF multiplier does not rely at all on increases in exports, since there is only a very modest fall in the real exchange rate, but is instead largely based on consumption, which grows much faster than the real disposable income measure, adjusted for the inflation loss on liquid assets, which is used in the OEF consumption equations. An important explanation for this strong consumption response is that, in the modelling of the direct influence on consumers' expenditure on non-durable goods and services of the real interest rate, its two components, namely the nominal interest rate and the inflation rate, are entered separately, and the positive coefficient on inflation is greater than that required to define the real interest rate, given the coefficient on the nominal interest rate. This excessive responsiveness to inflation has no counterpart in the other models.

The flat HMT multiplier response masks considerable differences in the response of the various expenditure categories. For example there is a large increase in investment, especially in non-manufacturing, which grows much faster than output in the short term although there is a unit long-run output elasticity in this equation. In contrast to the other models, however, exports fall, as a result of the slight appreciation of the real exchange rate, discussed above. This also makes a small contribution to the strong import growth, which principally arises from the large long-run elasticity between imports of manufactures and the demand for manufactures of about 1.5.

To the extent that the increase in government expenditure simply represents a shock to demand and so does not change the NAIRU, the resulting fall in unemployment, the lower real exchange rate (except HMT), and increased capacity utilisation (especially LBS and OEF) should all combine to produce increasing inflation over the course of the simulation horizon. This is indeed the case for the LBS, HMT and BE simulations, but not for OEF. Inflation in the OEF model stabilises towards the end of the simulation at a rate which implies that by year 5 the inflationary cost of reducing unemployment (as measured by year-on-year change in the price level normalised on the fall in unemployment) is lower than for the other quarterly models. This result occurs, despite increased inflationary pressure in both the labour market and the goods market, because the absence of static price homogeneity throughout the wage-price-exchange-rate sectors severely dampens the transmission of inflation. Conversely the inflationary cost of reducing unemployment by year 5 is greatest for the LBS and BE models. In the LBS model this results from the very strong demand-pull pressures in the labour market and from the persistence of higher capacity utilisation, discussed above. However over the early years of the simulation this inflationary cost is the smallest of any of the models, owing to the sluggishness of price adjustment, also discussed in Section 2.

In the BE model, demand-pull inflationary pressures from the labour market are relatively weak (see table 1), and a more important source of inflationary pressure is the fall in the real exchange rate, which depreciates both because of a deterioration in the current account and because of an increase in the PSBR. This movement in the real exchange rate contributes roughly half of our measure of the inflationary cost of reducing unemployment. Our ranking of the BE simulation as having a high inflationary cost may however be misleading, as the relationship between changes in employment and changes in unemployment is significantly different from that on the other models: if the change in GDP is instead used to normalise the inflation then the BE simulation is no more inflationary than HMT or OEF. According to the BE equation determining unemployment, whereas additional jobs in manufacturing reduce unemployment on a 1:1 basis, four times as many jobs in non-manufacturing including government sector) are required to achieve the same fall in registered unemployment. While there is broad agreement across the models (with the exception of LBS) that manufacturing jobs are more effective at reducing unemployment than non-manufacturing jobs, the size of this differential effect continues to be a cause for concern. In this respect recent work at NIESR, who also model changes in unemployment rather than labour force participation in terms of a behavioural equation (Turner et al., 1989, discuss these alternative approaches in relation to an earlier vintage of models) suggests that further disaggregation of the non-manufacturing sector as well as distinguishing between the effect of part-time/fulltime and male/female employment on unemployment produces results which are more intuitively appealing (Gregg, 1989).

In the HMT simulation inflation largely originates in the labour market, with upward pressure on real wage costs not only from lower unemployment but also increased company profitability. Increased capacity utilisation significantly contributes to inflation only at the beginning of the simulation and the rise in the real exchange rate (although implausible) acts to dampen the increase in import prices. Tax cut simulations In order to explain the simulation results of cuts in taxation, we begin by examining their direct effect on two measures of the real wage, namely the workers' real consumption wage and employers' real wage costs, since these concepts are useful in understanding the transmission mechanism through which tax cuts work. To the extent that it increases workers' real consumption wages the tax cut operates through increased demand via higher disposable income, with similar results to the demand shock discussed above, namely an increase in government expenditure. Conversely to the extent that it reduces employers' real wage costs the tax cut acts as a supply shock, lowering the NAIRU by reducing pressure from the push variables' which drive the inflationary process. As we show below the wage equation is critical in determining the incidence of the tax cut on employers and employees.

The dependence of the long-run solution to the model wage equations on taxes can be represented as

(1) w = [p.sup.C] = [dela.sub.1.t.sup.d] [dela.sub.2.t.sup.i] - [dela.sub.3.t.sup.e] + other terms,

0 [less than or equal to] [dela.sub.1, dela.sub.2, dela.sub.3 [less than or equal to] 1.

where w is the log of nominal earnings, [p.sup.C] is the log of consumer (or retail) prices and [t.sup.d, t.sup.i and t.sup.e] are the average direct, indirect and employers' tax rates (expressed in percentage form) respectively. The workers' real consumption wage is defined as take-home (net of direct taxes) average earnings deflated by consumer prices, or in log terms (w-[p.sup.C]-[t.sup.d].

Then substituting from (1) gives

(2) Workers' real consumption wage (log)

= w - [p.sup.c] - [t.sup.d] = ([dela.sub.1]-1)[t.sup.d] [dela.sub.2.t.sup.i] - [dela.sub.3.t.sup.e +...]

Employers' real wage costs are average earnings plus employers' contributions deflated by producer prices, or in log terms (w-[p.sup.p]+ [t.sup.e]), where pp is the log of producer prices. if we assume for expositional purposes that consumer prices are simply producer prices plus a mark-up for indirect taxes (ignoring, for example, import prices) so that [p.sup.c] = [p.sup.p] + [t.sup.i] then substituting from (1) gives

(3) Employers' real wage costs (log) = w - pp + t'

= W - (p.sup.c] - [t.sup.i]) + [t.sup.e]

= [dela.sub.1.t.sup.d] + (1 - [dela.sub.2])t.sup.i + (1-[dela.sub.3])[t.sup.e] +...

Although in full model simulation other variables such as unemployment impinge on wages we can use equations (2) and (3) to derive the direct effect of a cut in any of the tax rates on the workers' real consumption wage and employers' real wage costs. For example, considering a fall in income taxes M, if none of this cut is absorbed in lower nominal wages al = 0) then the tax cut is completely reflected in a rise in the real consumption wage and then operates as a shock to demand. Similar results arise for [dela.sub.2] =1 or [dela.sub.3] = 1, for cuts in [t.sup.i] and [t.sup.e]. At the other extreme, if [dela.sub.] = 1 then employers' real wage costs fall by the full extent of the income tax cut, hence lowering the NAIRU. Again identical results occur with [dela.sub.2] = 0 or [dela.sub.3] = 0 for a fall in [t.sup.i] and [t.sup.e] In practice the only non-zero values of the x-coefficients in the quarterly models' wage equations are as follows: for HMT [dela.sub.1] = 0.6; for BE [dela.sub.1] = 0.2 (on average allowing for interaction between wage sectors); and for NIESR [dela.sub.1] = 0.47, 12 = 0.53 and [dela.sub.3] = 0.53. A notable feature of the NIESR coefficients is the imposition of the restriction (1 - [dela.sub.1]) = [dela.sub.2] = [dela.sub.3], which implies that the long-run effects of equivalent changes in the tax rates [t.sup.d, t.sup.e] and [t.sup.e] are identical. However it is also worth noting that NIESR themselves concede that the 'estimated coefficients on individual tax effects were often not robust'. Income tax cut Differences in the size of the retention ratio effect, [dela.sub.1] equation (1), continue to be an important explanation of differences in this simulation. The retention ratio is the ratio of the net wage to the gross wage where the former is defined as gross earnings less income tax and employees' national insurance contributions. This variable is an important determinant of wages if it is assumed that workers bargain for a target real consumption wage.

Given that there is no long-run retention ratio effect in either the LBS or OEF models (although in the latter case there is some temporary short-run effect) much of the initial stimulus to GDP therefore comes from higher consumption as a result of increased disposable income. In both cases this is also reinforced by similar mechanisms to those described for the government expenditure simulation. The absence of any long-run retention ratio effect means that by years 4-5 the inflation-unemployment tradeoff in the LBS and OEF simulations is similar to that under the government expenditure simulation, perhaps not a surprising result given that both policy changes can simply be regarded as a shock to aggregate demand, although increased government expenditure delivers a much more rapid fall in unemployment. In the OEF simulation a notable feature of the short-term results is that there is a complete absence of any rise in prices over the first two years, a result which is explained by a powerful short-run retention ratio effect, although as can be seen by the subsequent response of real wages, this seemingly has little influence on the longer-run simulation results.

In the BE simulation there is a small long-run retention ratio effect, [dela.sub.1] = 0.2, and this entails a fall in unemployment (and rise in GDP) similar to that in the government expenditure simulation by year five, but with an increase in the consumer price level of only two-thirds that in the government expenditure case. The influence of a much larger retention ratio effect ([dela.sub.1] = 0.6) can be seen more clearly in the HMT simulation, where there is a more substantial fall in employers' real wage costs and hence greater downward pressure on consumer prices. By the fifth year of the simulation the rise in the consumer price level is only about one-fifth of the rise following the government expenditure increase, but unemployment has fallen by slightly more than in the previous simulation.

The influence of the retention ratio effect is less immediately discernible in the NIESR simulation because the domestic price level still rises in response to the step fall in the nominal exchange rate. Nevertheless employers' real wage costs do fall and hence put downward pressure on inflation thus allowing a long-run fall in the NAIRU. By the fifth year of the NIESR simulation the model appears to be converging on this new NAIRU with an overall fall in unemployment of about 40 thousand, similar to that achieved in the government expenditure simulation. There are some important differences between the two simulations, however. For example, because the present result does not rely on the appreciation of the real exchange rate that occurs in the government expenditure simulation, there is an improvement rather than a decline in manufacturing output. A cut in VAT. The direct effect of a cut in VAT by one percentage point is to reduce the price level by less than 1 per cent in the NIESR, BE, HMT and OEF models (reflecting the fact that VAT does not cover all goods and services). In the case of LBS the single-equation response of the consumer prices equation implies that a change in VAT only has a temporary effect on the price level. The cause of this counter-intuitive property can be traced to the functional form in which VAT enters this equation (also highlighted in our previous review of model properties). A re-estimate of the LBS equation using the NIESR measure of indirect taxes in a more suitable functional form allows the imposition of a restriction which implies that changes in indirect tax rates have, allowing for coverage, a one-for-one effect on consumer prices. This feature of the original LBS equation makes it difficult to interpret the VAT simulation, although it would seem that the effects of a cut in VAT are very similar to those of a cut in income tax.

The absence of any indirect tax effects in the HMT, BE and OEF wage equations ([dela.sub.2] = 0) means that the direct effect of the cut in VAT is largely reflected in a fall in employers' real wage costs. This reduction is then passed on in lower prices which in turn leads to an appreciation of the exchange rate and further downward pressure on prices and so on, so that beyond the first year the fall in the price level is significantly more than the direct effect of the cut in VAT. This downward spiral in prices is counteracted in the BE simulation by downward pressure on the real exchange rate resulting from the deterioration in the current balance and the increase in the PSBR; and in the HMT and OEF simulations by the lack of homogeneity in the exchange-rate equation. A further limiting factor which is common to these three models is the upward pressure on real wages from both labour market and goods market sources. These counteracting inflationary pressures eventually retard the fall in the price level in the HMT and OEF simulations, although the fall in the price level in the BE simulation shows no sign of slowing.

The effect of a cut in VAT on the NIESR model is very similar to a cut in the basic rate of income tax particularly in the long run, a result which is consistent with the above analysis, the discrepancy being due to differences in the implied change in average tax rates in the simulations. In the short run the VAT cut has some advantages over the income tax cut from its more immediate deflationary impact on the price level, so that the initial increases in the price level are slightly less in the VAT simulation. A cut in employers' NICs. Given that the coefficients on employers' contributions in the model wage equations are identical to those on indirect taxes in the HMT, NIESR, BE and OEF models (i.e. ot, = oc,) it is not surprising that a cut in employers' NiCs has very similar effects to a cut in VAT in all these models, although it should be borne in mind that the magnitude of the two tax changes are not identical. The main difference would seem to be that in all four mode(s the cut in VAT has a much more immediate impact on the price level, whereas additional lags follow the cut in employers' NICs before lower labour costs are passed on in lower wholesale and consumer prices. An additional reason why a cut in employers' NICs has a less immediate impact is that both the HMT and BE wage equations imply that the cut in NiCs leads to a temporary increase in wage demands, as workers perceive that firms can afford higher pay increases, although there is no such permanent effect (i.e. although [dela.sub.3] = 0, in the short run there is a positive coefficient).

In the LBS simulation the effect which the cut in employers' NICs has on lowering wage costs is substantially outweighed by the inflationary effect of the fall in the exchange rate, so that consumer prices rise (as is the case for NIESR). However it is puzzling that the inflationary effects in this simulation are at least as great, when normalised on either GDP or unemployment, as for simulations of an income tax cut or an increase in government expenditure on this model. Problems with the specification of the wholesale price equation, namely sluggish adjustment, noted above, and the comparatively low weight which wage costs have in determining prices compared to other costs, contribute to this result. In addition, for no obvious reason, employers' NICs are excluded altogether from the definition of the wage cost variable which appears in the consumer price equation, even though they are included in the wage cost variable which appears in the wholesale price equation.

Cut in short-term nominal interest rates In this exercise we consider the effects of reducing short-term interest rates by 1 percentage point. Term structure equations in the models relating long rates to short rates are assumed to continue to operate. In the LPL model interest rates are jointly determined with the exchange rate to maintain uncovered interest-rate parity and hence this experiment is not performed for this model. The exchange rate in the NIESR model can be regarded as an uncovered interest-rate parity condition with the inclusion of a term in the ratio of the net acquisition of overseas assets to GDP, reflecting a risk premium. This produces the result that the initial fall in the exchange rate for a permanent change in interest rates depends on the length of the simulation period and hence can produce an excessively large initial reaction. In our simulation the shock lasts for the five years of the reporting period but it is assumed that expectations are formed on the basis that the shock does not persist thereafter and hence is regarded as temporary.

In examining the transmission mechanism in the quarterly models through which interest rates operate on consumption, investment and the exchange rate, we distinguish between direct and indirect effects. The direct effects are the single-equation partial responses, and the full simulation results (table A5) are then the sum of these direct effects plus indirect effects which occur from feedback through all the model equations. The direct effects on consumption range from 0.32 per cent for LBS to 0.65 per cent for BE, although in the latter case there are very long lags, the median lag for the effect of interest rates on non-durable consumption being about four years. There is general agreement across the models that, proportionately, the largest direct effects are on private housing investment, ranging from 1 per cent for OEF to 3.3 per cent for BE. However while all of the models, with the exception of LBS, have significant direct effects on manufacturing investment, there is much greater disagreement on the relative effects on non-manufacturing investment. In those models which make little attempt to disaggregate the non-manufacturing sector (LBS, HMT and OEF) there are no long-run direct interest rate effects, whereas those models which do disaggregate the non-manufacturing sector (NIESR and BE) find significant direct interest-rate effects. It is the latter models where, not surprisingly given the importance of the non-manufacturing sector in determining total investment, the overall direct effects of interest rates are clearly larger on investment than on consumption. This result is particularly true for the BE model where a large interest-rate effect on distribution and other services' contributes to the result that every 1 percentage point fall in interest rates increases total fixed investment by about 1 per cent.

There are also substantial differences in the response of the exchange rate to interest rates. The direct effect of every 1 percentage point cut in interest rates is to lower the real exchange rate by 1.3, 0.6 and 0.5 per cent according to the HMT, BE and OEF models respectively. In the NIESR model a fall in interest rates of 1 percentage point which is expected to last for 5 years leads to an approximate fall in the nominal exchange rate of about 5 per cent, with the exchange rate subsequently appreciating by about 1 per cent per annum. Given that the exchange rate is determined by the complete interaction of the LBS financial sector it is not possible to calculate the direct effect of a cut in interest rates for this model. As previously argued this can be a serious limitation to understanding full model simulation properties, a point which is particularly relevant in the present case where beyond the first year the cut in interest rates produces the counter-intuitive result of an appreciation in both the nominal and the real exchange rate. The rise in the real exchange rate, together with comparatively weak direct effects on consumption and investment, explains why the GDP effects are so small in the LBS simulation.

The NIESR simulation results are also dominated by the response of the exchange rate. Thus in the first year of the simulation the large fall in the nominal exchange rate improves competitiveness and leads to the largest GDP response of any of the models. However, as the nominal exchange rate subsequently rises and prices increase, the improvement in competitiveness and hence the GDP response are dampened.

In the HMT simulation the overall stimulus to GDP is very much larger than that suggested by the direct effects on consumption and investment. This is partly explained by the fall in the real exchange rate, although the absence of price homogeneity in the exchange-rate equation means that this fall is not as large as the direct effect. in addition there is a further stimulus to consumption from the effect which interest rates have on improving real wealth, and by the fourth year the rise in consumption is nearly 9 per cent. The total effects on fixed investment are also much larger than the direct effects, particularly due to a very large accelerator response from non-manufacturing investment: beyond the third year the total rise in fixed investment is more than 3 per cent. These accelerator effects on fixed investment operate even more strongly in the BE simulation so that in combination with large direct interest-rate effects they produce a rise in fixed investment of about 4 per cent by year 5, which is nearly three times the percentage increase in consumption. Given the absence from the OEF model of any large direct interest-rate effects on either consumption, investment or the exchange rate, the overall GDP response in this simulation is significantly less than for either HMT or BE, with a more balanced increase (in percentage terms) in the expenditure components.

4. Conclusion

There have been important developments in the supply side of the main large-scale macro models, and these have changed their overall properties, as revealed in standard simulation experiments. In particular, expansionary demand policies now tend to be associated with smaller increases in domestic output in the longer run than in earlier vintages of these models (see, for example, Wallis et aL, 1986). The principal cause of this change is the greater degree of inflationary pressure generated by a demand expansion, and the subsequent feedback effects of higher inflation in reducing demand. In the short run, however, the quarterly models appear very Keynesian. For example, although the NIESR model can no longer be classified as Keynesian in respect of its simulation responses at the end of a five-year period, it exhibits these properties in the short term because wages and prices adjust relatively slowly despite rational expectations in wage and price setting.

Greater emphasis on the supply side and the inflation mechanism in the quarterly models is a response to the increased focus on these issues in policy making. It can also be seen as a response to the challenge of the LPL model, first used for forecasting in 1980, and placing great emphasis on the supply side of the economy. The overall specification of the LPL model relies heavily on predefined economic theory, thus many of its properties are known a priori, and these have remained largely unchanged since the model's inception. It is an equilibrium model which assumes market clearing in all sectors of the economy and corresponds to the new classical paradigm. it remains an outlier, however, many of its properties being at variance with those of the quarterly models which, despite giving greater emphasis to the supply side, reflect a stronger reliance on empirical evidence, and in particular the view that many markets do not correspond to the perfectly competitive paradigm. Simple monetarist models of inflation, prevalent in the early 1980s, have been proved inadequate, and the quarterly models now attempt to provide fully elaborated representations of the inflation mechanism.

Differences in the long-run properties of the quarterly models can be largely explained by differences in the key wage-price-exchange-rate interactions. in turn these differences can often be resolved by examining the economic and statistical credentials of these structural relationships. However they also arise because different groups attach different relative weights to economic theory and empirical evidence in the detailed specification of the model, and two examples stand out in this paper. The first is the issue of homogeneity in the wage, price and exchange-rate equations. Some models, for example NIESR and HMT, place a high weight on the theoretical requirement of homogeneity, whereas the OEF model relies more on statistical estimates of the relevant parameters and does not impose the corresponding homogeneity restrictions. This difference of emphasis has important implications for overall model properties, and specifically for the nature of the long-run response in the OEF model.

The second example concerns the role of taxes. Our analysis in Section 3 shows that estimates of the response of the economy to a change in tax rates, both between models and between different types of taxes, can be explained largely by inspection of the parameters of the wage relationships. These tax coefficients tend to be poorly determined in empirical estimation, however, partly because the aggregate tax variables do not adequately reflect changes in the particular taxes which influence decisions at the micro level. In response to this uncertainty some modellers, specifically NIESR, choose to impose the theoretical prior that the long-run effect of different tax changes on wages, and hence the NAIRU, should be equal, whereas others rely on statistical evidence, which often fails to indicate that a significant non-zero effect is present. Again this difference has important implications for model-based policy analysis, and its resolution may require a more disaggregated treatment.

Of the quarterly models considered in this paper only that of NIESR can be said to possess a clearly defined long run, although all the models have incorporated more supply-side factors in their structure. Further research is required before it can be determined whether the current divergencies from the NIESR framework observed in the other models represent genuine differences in approach or differences in detailed specification that can be empirically resolved. Despite differences in the treatment of the supply side the models nevertheless represent a useful common framework within which to interpret the inflationary processes in the economy and, more generally, through ready-reckoners, to provide a quantitative assessment of policy options.

REFERENCES

Courakis, A.S. (1988), Modelling portfolio selection', Economic Journal, 98, pp. 619-641. Fisher, P.G. and Turner, D.S. (1990), The exchange rate, forward expectations and the properties of macroeconometric 'models', Discussion Paper No.23, ESRC Macroeconomic Modelling Bureau. Fisher, P.G., Tanna, S.K., Turner, D.S., Wallis, K. F. and Whitley, J.D. (1989), Comparative properties of models of the UK 'economy', National Institute Economic Review, No. 129, pp. 69-87. Fisher, P.G., Tanna, S. K., Turner, D. S., Wallis, K. F. and Whitley, J. D. (1 990), 'Econometric evaluation of the exchange rate

In models of the UK economy', Economic Journal, forthcoming. Gregg, P. (1989), 'Out for the count-a new approach to modelling claimant unemployment', Discussion Paper No. 167,

National Institute of Economic and Social Research. Joyce, M. and Wren-Lewis, S. (1989), 'Does the NAIRU matter?', Discussion Paper No.1 54, National Institute of Economic

and Social Research. Layard, P.R.G. and Nickell, S.J. (1985).'The causes of British unemployment', National Institute Economic Review, No.1 1 1,

pp. 62-85. Melliss, C.L. (1988), 'HM Treasury macroeconomic model 1986', Economic Modelling, 5, pp. 237-259. Melliss, C., Meen, G., Pain, N. and Whittaker, R. 1989),'The newtreasury model project', GES Working Paper No.106. Turner, D.S., Wallis, K.F. and Whitley, J.D. (1989), Differences in the properties of large-scale macroeconometric models:

the role of labour market specifications', Journal of Applied Econometrics, 4, pp. 31 7-344. Wallis, K.F. (ed.), Andrews, M.J., Fisher, P.G., Long bottom, J.A. and Whitley, J.D. (1986), Models of the UK Economy: A

Third Review by the ESRC Macroeconomic Modelling Bureau, Oxford: Oxford University Press. Wren-Lewis, S. (1989), The macroeconomics of supply', Discussion Paper No.160, National institute of Economic and

Social Research.

The nature of the association between inflation and the level of unemployment has been a persistent issue of controversy over the last three decades. initially, attention focussed on the statistical relationship between nominal wage inflation and unemployment-the Phillips curve-which could be seen equally as a relationship between price inflation and unemployment, if prices are a constant mark-up on wages. This was quickly adopted as a menu for policy choice, describing the trade-off between increases in unemployment and reductions in inflation. By the 1970s, however, the question was whether a long-run trade-off existed at all, the OECD economies having experienced rising unemployment and, simultaneously, rising inflation. The subsequent re-examination of labour market behaviour introduced the concept of an equilibrium rate (the natural rate) of unemployment which, in the monetarist view, was not amenable to demand management policies. More recent developments reflect a growing concern with the supply side of the economy, including the question of what determines the non accelerating inflation rate of unemployment (NAIRU).

The emerging framework uses bargaining models for the determination of wages: unionised employees set pay demands and employers respond according to their ability to meet these demands, depending in turn on their pricing decisions. In an influential analysis, Layard and Nickell (1985) develop a three equation model, determining wages, prices and employment, and emphasise the role of 'push' variables, which can affect firms' price mark-up on wage costs or wage bargainers' mark-up on prices, that is, the real wage. The push variables include taxes and benefits, import prices, and productivity. For given values of such variables there is a unique level of unemployment, the NAIRU, which leads bargainers to settle for a real wage which is consistent with that which firms are willing to accept in their pricing behaviour. Large-scale macroeconometric models differ substantially in size from the Layard-Nickell model, nevertheless this provides a useful starting point in studying their properties, given that they increasingly incorporate aggregate supply considerations. Extensions to the Layard-Nickell framework are necessary, however, most notably to take account of the endogeneity of the real exchange rate, which is an important adjustment mechanism in an open economy with floating exchange rates. Additional push factors, such as a deterioration in the current balance, can then act to produce higher relative prices of imports, in turn increasing the NAIRU. Recent NIESR research (Joyce and Wren-Lewis, 1989; Wren-Lewis, 1989) shows how the change in the NAIRU following any policy change on the NIESR model can be derived as a simple function of the long-run solutions to the model's wage and price equations, together with the response of the real exchange rate.

This paper describes six major macroeconometric models of the UK economy, using their wage-price-exchange-rate interaction as a core supply-side framework in which to interpret their properties, as revealed in standard simulation experiments. The wage, price, and exchange-rate equations provide considerable insight into the simulation results, and these are accordingly analysed next, in Section 2. Some general features of the simulations are noted in the course of this discussion, with detailed analysis presented in Section 3. The numerical results of the policy simulations for the major macroeconomic indicators-ready reckoners-are presented in Appendix tables A1-A5, and Section 4 contains concluding comments. The models considered are those of the London Business School (LBS), National institute of Economic and Social Research (NIESR), Her Majesty's Treasury (HMT), Bank of England (BE), Oxford Economic Forecasting (OEF) and the Liverpool University Research Group in Macroeconomics (LPL), as deposited with the Bureau in late 1989; the HMT model was also publicly released at this time. All of these are quarterly models, except the LPL model, which is annual, and as this has remained largely unchanged over the last six years there is little discussion of it in the present paper: reference may be made to our earlier analyses in this Review and elsewhere (for example, Fisher et aL, 1989). The OEF model is a newcomer to our comparative analysis. 2. Wages, prices and the exchange rate Differences in the way in which wages, prices and the exchange rate are determined in the various models provide a major part of the explanation of differences in their overall properties. This section therefore presents an analysis of the relevant equations, as a prerequisite to an understanding of the policy responses.

Wage equations

The wage equation in most macro models is an important link in the resolution of the overall association between changes in unemployment and inflation. Even at the single-equation level there is wide diversity across the models in the long-run response of wages to a change in unemployment, as shown in table 1. The functional form in which unemployment enters the wage equation differs across the models, so for comparative purposes we examine the sensitivity of wages to a fall in unemployment of 100 thousand from an initial level of 2 million, allowing where relevant for interaction between wages in different sectors (LBS and BE) or for changes in the duration structure of unemployment (LBS; BE and NIESR have now removed this distinction in the determination of wages). The LBS model is a clear outlier, with the rise in total wages nearly three times greater than that in any other model. This is in turn reflected in the full model simulation results, as the rise in real wages is much greater in all the LBS simulations, by the last year. The sector which is most responsible for these effects is manufacturing, whose wages then determine those in non-manufacturing. The LBS manufacturing wage equations are very different not only from comparable equations in the other models but also from wage equations in previous vintages of the LBS model, and econometric analysis reveals severe problems; of particular importance is the instability of the estimate of the long-run real wage response to a change in aggregate unemployment. The BE wage equations by contrast have a particularly low response to changes in unemployment, but this key response is ill-determined statistically.

The functional form in which unemployment appears in the wage equation is one of the major sources of non-linearity in the models, and must be borne in mind before scaling up the ready reckoner results to evaluate much larger policy changes. in the LBS, HMT, OEF and LPL wage equations the logarithm of the unemployment rate is used, implying that the same absolute fall of 1 00 thousand from an initial level of unemployment of one million would double all the wage effects in table 1. in the BE wage equations the reciprocal of the unemployment rate is used implying, approximately, a quadrupling of the wage responses in table 1 if the initial level of unemployment is reduced to 1 million. The NIESR model is an interesting exception since, rather than the official claimant measure of unemployment, the proportion of the adult population not in work is used in the wage equation, on the grounds that this is a superior measure of wage pressure, and as this variable is not transformed the NIESR response in table 1 is not base dependent.

The single-equation response of nominal wages to changes in consumer/retail prices is very rapid across all of the quarterly models, with nominal wages rising more or less in line with prices. An important exception is the OEF wage equation, where the elasticity of nominal wages with respect to prices takes a value less than one, namely the unrestricted point estimate of 0.93.

All of the wage equations include a time trend or some measure of productivity to explain the growth in real wages over time. Of particular importance is the use of a direct measure of productivity in the NIESR model: as increases in employment tend to lag behind increases in output a sharp rise in productivity results, contributing to pronounced increases in real wages in the first two years of all the NIESR simulations, although real wages subsequently fall back as the employment response comes through. A further contribution to these short-term surges in real wages comes from a 'mismatch' variable which has a strong influence whenever employment is changing rapidly. Such sharp increases in real wages do not feature in the other model simulations since those models which include productivity in the wage equation (HMT, BE and OEF) use a trend or smoothed measure of productivity, and in no other model does a mismatch' variable influence wages. The response of real wages helps to explain (in conjunction with the exchange-rate response) why the initial effects on both inflation and demand tend to be larger in the NIESR simulations than for other models.

Price equations

In all of the quarterly models, prices are determined largely as a mark-up on costs, with a 1 per cent increase in all costs (mainly labour and import costs) eventually leading to a 1 per cent increase in prices. However for a number of the equations in the OEF price sector this condition of static homogeneity has not been imposed, even though in most cases the freely estimated coefficients do not depart substantially from this restriction. This feature of the OEF price equations together with the absence of static homogeneity in the wage and exchange-rate equations seriously dampens the transmission of inflationary pressures in this model. it is a consequence of a modelling strategy in which empirical validity has received higher priority than theoretical consistency.

The equation for manufacturing wholesale prices can be regarded as the pivotal equation in all cases, although the extent of disaggregation of the price system differs significantly across the quarterly models, as does the speed of adjustment. According to the NIESR, BE and HMT wholesale price equations, 50-60 per cent of a step increase in total costs is passed on in higher prices within two quarters, and at least 90 per cent within four quarters. The response of the OEF equation lags approximately one quarter behind these models, probably because no contemporaneous effect of costs on wholesale prices is allowed. However the main outlier is the LBS equation, since it takes six quarters before 50 per cent of an increase in total costs is passed on in higher prices and more than four years before this proportion reaches 90 per cent. This sluggishness is likely to dampen substantially any inflationary effects in full model simulations. Our re-estimation of the LBS equation suggests that these responses are sensitive to the sample period and that if this is confined to the more recent data then the responses are very similar to those from the other models.

In all of the manufacturing wholesale price equations the size of the mark-up on costs is sensitive to the pressure of demand through some measure of capacity utilisation. However from the point of view of full model simulations the way in which capacity utilisation is determined is perhaps of greater importance than the single-equation effect of any given change in capacity utilisation on prices. Thus, in the BE and HMT models capacity utilisation is basically determined as a distributed lag on manufacturing output so that any increase in output, once it stabilises at a constant difference from base, eventually leaves capacity utilisation unchanged. On the other hand, in the LBS, NIESR and OEF models, changes in capacity utilisation emerge from the dynamics of output and capital stock adjustment, as determined elsewhere in the model. In the case of NIESR this adjustment process is very prolonged and usually cyclical but does seem to ensure that capacity changes in line with output in the long run. However this does not appear to be the case for either the LBS or OEF models, where expansionary policies are associated with persistently higher capacity utilisation, which might seem to be a doubtful long-run property. By itself this would also result in inflationary effects that are stronger than in other models; that these are not observed is principally due to features discussed above, namely the very sluggish price dynamics in the LBS model, and the absence of static homogeneity throughout the OEF wage-price sector and in its exchange-rate equation, discussed next.

The exchange rate

In explaining the role of the exchange rate it is useful to assess the models' exchange-rate equations according to three characteristics: the extent to which they are forward-looking; their responsiveness to changes in competitiveness; and their responsiveness to current account imbalance.

The exchange rate in the LBS and NIESR models is forward-looking and hence in response to a shock the nominal exchange rate immediately 'jumps' part of the way (LBS), or almost all of the way (NIESR) to the long-run change in the exchange rate. The large fall in the exchange rate on the introduction of expansionary policies means that inflationary pressure from higher import prices and the stimulus to GDP via improved competitiveness and hence net exports are greater in the early years of the LBS and NIESR simulations in comparison with other models. The LBS exchange rate is determined as a market-clearing price in the financial sector of the model and hence it is not possible to analyse the precise mechanism leading to a change in this variable. Given the key role of the exchange rate in determining both short- and long-run model responses, this is a serious handicap to understanding the simulation properties of the LBS model, particularly given the criticism of many parameter values in its financial sector made by Courakis (1988). The long-run change in the NIESR exchange rate is determined so as to correct any current account imbalance, or more precisely any imbalance in net overseas assets: an ex ante deterioration in the current account balance equivalent to 1/2 per cent of GDP (about 22.2 billion in 1990 prices) leads to a step fall in the nominal exchange rate of roughly 6 per cent. Fisher and Turner (1990) show that the magnitude of this response is sensitive to those parts of the model which deal with the effect of changes in wealth, including revaluations; unfortunately these effects are among the most difficult to quantify.

The exchange rate in the BE model is specified in real terms so that in simulation the nominal exchange rate falls ceteris paribus in line with increases in (wholesale) domestic prices. This homogeneity property would appear to be one which HMT and OEF (which has an equation closely based on the HMT specification) have attempted to incorporate in their exchange-rate equations. Thus according to Melliss (1988, p.248) the HMT equation

'uses the idea that relative money supplies and relative

labour costs are an indicator of the future movements in

UK relative to world prices. The unit coefficient on the

sum of these terms in a log specification indicates that

the equation is for the equilibrium real exchange rate.'

[emphasis added]

However in the various simulations on the HMT and OEF models, relative money supplies are not associated one-to-one with future movements in prices, at least within the time period considered. Moreover the actual association varies between simulations, so that the real exchange rate may rise or fall depending on the nature of the shock. In the case of OEF the growth in broad money tends to exceed the change in prices to the extent that output rises, and so this typically contributes to a fall in the real exchange rate. However this effect is partially offset by the inclusion in the OEF exchange-rate equation of UK domestic labour costs specified in dollar rather than sterling terms. This means that following an increase in UK labour costs the exchange rate does not depreciate as it ought to do in order to maintain the homogeneity property.

The current account balance enters into the OEF and BE exchange-rate equations: following an ex ante 1 billion pounds (1990 prices) deterioration in the current balance there is a long-run fall in the exchange rate of about 0.6 per cent for OEF and about 0.2 per cent for BE. A measure related to the current balance is also included in the HMT exchange-rate equation, but its coefficient is so small that it has a negligible influence. in addition a (cumulative) measure of the PSBR is included in the BE equation, as a measure of a risk premium, so that a El billion per annum increase in the PSBR leads to a fall in the exchange rate of about 0.3 per cent after five years.

Our analysis of the wage, price and exchange-rate equations in the quarterly models has revealed a number of statistical deficiencies and theoretical inconsistencies, which except in the case of the NIESR model preclude a straight forward derivation of the models' overall long-run properties. Nevertheless an understanding of the wage-price-exchangerate interaction, including any shortcomings, provides useful guidance in comparing the results of standard simulations, to which we now turn.

3. Standard simulations

The policy simulations considered in this section comprise an increase in government current expenditure of E2 billion per annum (1990 prices) reductions in the rate of income tax, VAT and employers' national insurance contributions of one percentage point, and a cut in nominal interest rates, also of one percentage point. It is assumed that any increase in the PSBR is accommodated under constant nominal interest rates, approximating money finance, and that all changes are assumed to be unanticipated and permanent, unless otherwise stated. General results on the major macroeconomic indicators over a five-year horizon, the longest period common to all models, are presented in tables A1-A5.

Our discussion extends that of the previous section by considering, in particular, the links between inflation and demand, these being necessary if convergence to a NAIRU is to occur. In several cases we observe that there is no such tendency and we are then able to use the foregoing discussion of the wage, price and exchange-rate equations to describe the nature of the inflationary processes themselves.

Government expenditure

In this section we examine the effects of an increase in government current expenditure, where the increase is allocated proportionately between procurement and employment in those models that make this distinction (LBS, HMT and OEF) and between central and local government where this additional disaggregation occurs (HMT). The size of the government expenditure increase is El,440 million per annum in 1985 prices, a figure which has been chosen so as to produce an increase in 1990 prices of about 22 billion per annum.

The GDP multipliers (defined as the ratio of the absolute increase in GDP to the increase in government expenditure) are shown in table 2. For all the quarterly models, except NIESR, the impact multiplier in the first year is about unity. After five years the multiplier has risen to between 1.6 and 1.8 on the LBS, BE and OEF models, although much of the rise in the LBS multiplier takes place in the second year rather than gradually, as in the case of the other two models, whereas the HMT multiplier remains virtually at unity over the entire simulation horizon. The NIESR multiplier clearly shows the greatest volatility with an impact multiplier of 1.8 rising to 2.2 in the second year before failing back sharply to only 0.4 by year 5.

The increase in government expenditure on the NIESR model leads to a decline in the model's NAIRU and hence a long-run increase in GDP, because government expenditure is generally less import-intensive than private sector expenditure and so brings about an increase in the long-run real exchange rate. This in turn reduces inflation by making imports relatively cheaper and so enables a longrun fall in unemployment of about 30-40 thousand to take place without increasing overall inflationary pressure. However in the short run there is a much larger multiplier effect which follows from the effect on demand of the initial improvement in competitiveness due to the step fall in the nominal exchange rate, as well as the temporary boost to real wages, discussed above. The adjustment from the large initial multiplier to the smaller long-run multiplier which is consistent with the change in the NAIRU, takes place through the effect which higher inflation has in eroding competitiveness and crowding-out consumers' expenditure, and hence reducing demand.

Explaining the simulation results in terms of the change in the NAIRU is more difficult for the other models. Partly this is because, as previously shown in Section 2, there are a number of inconsistencies in their supply-side structure which prevent calculation of the change in the NAIRU. However a second reason is that although the NAIRU defines an equilibrium in terms of inflation and unemployment there is no guarantee that the simulation will converge on this equilibrium. The NIESR model is an outlier in this respect: as noted above and in our previous review (Fisher et al., 1989, p.73), this model rapidly converges on the NAIRU. The absence of such strong adjustment mechanisms means that for the other quarterly models it is more appropriate to explain the response of unemployment and GDP in terms of changes in the components of demand. Our supplyside analysis in Section 2 is then useful in understanding the inflationary consequences of this change in demand.

The increases in the multiplier over the course of the BE and LBS simulations largely originate from the effect on demand of the fall in the real exchange rate, as is demonstrated by the fact that variant simulations in which the real exchange rate is held at its base values deliver a multiplier response which remains close to unity. The resulting improvements in competitiveness help to restrain imports and stimulate exports, which in turn boosts both stock-building and investment via accelerator type mechanisms. These accelerator effects are particularly strong in the BE simulation, with the absolute change in fixed investment nearly as large as that for consumption. This result is explained by the equations for manufacturing and non-manufacturing investment, which have elasticities with respect to sectoral output of 1.5 and 2.1 respectively, although the corresponding elasticities in the other models are typically constrained to take a value of unity.

The improvement in the OEF multiplier does not rely at all on increases in exports, since there is only a very modest fall in the real exchange rate, but is instead largely based on consumption, which grows much faster than the real disposable income measure, adjusted for the inflation loss on liquid assets, which is used in the OEF consumption equations. An important explanation for this strong consumption response is that, in the modelling of the direct influence on consumers' expenditure on non-durable goods and services of the real interest rate, its two components, namely the nominal interest rate and the inflation rate, are entered separately, and the positive coefficient on inflation is greater than that required to define the real interest rate, given the coefficient on the nominal interest rate. This excessive responsiveness to inflation has no counterpart in the other models.

The flat HMT multiplier response masks considerable differences in the response of the various expenditure categories. For example there is a large increase in investment, especially in non-manufacturing, which grows much faster than output in the short term although there is a unit long-run output elasticity in this equation. In contrast to the other models, however, exports fall, as a result of the slight appreciation of the real exchange rate, discussed above. This also makes a small contribution to the strong import growth, which principally arises from the large long-run elasticity between imports of manufactures and the demand for manufactures of about 1.5.

To the extent that the increase in government expenditure simply represents a shock to demand and so does not change the NAIRU, the resulting fall in unemployment, the lower real exchange rate (except HMT), and increased capacity utilisation (especially LBS and OEF) should all combine to produce increasing inflation over the course of the simulation horizon. This is indeed the case for the LBS, HMT and BE simulations, but not for OEF. Inflation in the OEF model stabilises towards the end of the simulation at a rate which implies that by year 5 the inflationary cost of reducing unemployment (as measured by year-on-year change in the price level normalised on the fall in unemployment) is lower than for the other quarterly models. This result occurs, despite increased inflationary pressure in both the labour market and the goods market, because the absence of static price homogeneity throughout the wage-price-exchange-rate sectors severely dampens the transmission of inflation. Conversely the inflationary cost of reducing unemployment by year 5 is greatest for the LBS and BE models. In the LBS model this results from the very strong demand-pull pressures in the labour market and from the persistence of higher capacity utilisation, discussed above. However over the early years of the simulation this inflationary cost is the smallest of any of the models, owing to the sluggishness of price adjustment, also discussed in Section 2.

In the BE model, demand-pull inflationary pressures from the labour market are relatively weak (see table 1), and a more important source of inflationary pressure is the fall in the real exchange rate, which depreciates both because of a deterioration in the current account and because of an increase in the PSBR. This movement in the real exchange rate contributes roughly half of our measure of the inflationary cost of reducing unemployment. Our ranking of the BE simulation as having a high inflationary cost may however be misleading, as the relationship between changes in employment and changes in unemployment is significantly different from that on the other models: if the change in GDP is instead used to normalise the inflation then the BE simulation is no more inflationary than HMT or OEF. According to the BE equation determining unemployment, whereas additional jobs in manufacturing reduce unemployment on a 1:1 basis, four times as many jobs in non-manufacturing including government sector) are required to achieve the same fall in registered unemployment. While there is broad agreement across the models (with the exception of LBS) that manufacturing jobs are more effective at reducing unemployment than non-manufacturing jobs, the size of this differential effect continues to be a cause for concern. In this respect recent work at NIESR, who also model changes in unemployment rather than labour force participation in terms of a behavioural equation (Turner et al., 1989, discuss these alternative approaches in relation to an earlier vintage of models) suggests that further disaggregation of the non-manufacturing sector as well as distinguishing between the effect of part-time/fulltime and male/female employment on unemployment produces results which are more intuitively appealing (Gregg, 1989).

In the HMT simulation inflation largely originates in the labour market, with upward pressure on real wage costs not only from lower unemployment but also increased company profitability. Increased capacity utilisation significantly contributes to inflation only at the beginning of the simulation and the rise in the real exchange rate (although implausible) acts to dampen the increase in import prices. Tax cut simulations In order to explain the simulation results of cuts in taxation, we begin by examining their direct effect on two measures of the real wage, namely the workers' real consumption wage and employers' real wage costs, since these concepts are useful in understanding the transmission mechanism through which tax cuts work. To the extent that it increases workers' real consumption wages the tax cut operates through increased demand via higher disposable income, with similar results to the demand shock discussed above, namely an increase in government expenditure. Conversely to the extent that it reduces employers' real wage costs the tax cut acts as a supply shock, lowering the NAIRU by reducing pressure from the push variables' which drive the inflationary process. As we show below the wage equation is critical in determining the incidence of the tax cut on employers and employees.

The dependence of the long-run solution to the model wage equations on taxes can be represented as

(1) w = [p.sup.C] = [dela.sub.1.t.sup.d] [dela.sub.2.t.sup.i] - [dela.sub.3.t.sup.e] + other terms,

0 [less than or equal to] [dela.sub.1, dela.sub.2, dela.sub.3 [less than or equal to] 1.

where w is the log of nominal earnings, [p.sup.C] is the log of consumer (or retail) prices and [t.sup.d, t.sup.i and t.sup.e] are the average direct, indirect and employers' tax rates (expressed in percentage form) respectively. The workers' real consumption wage is defined as take-home (net of direct taxes) average earnings deflated by consumer prices, or in log terms (w-[p.sup.C]-[t.sup.d].

Then substituting from (1) gives

(2) Workers' real consumption wage (log)

= w - [p.sup.c] - [t.sup.d] = ([dela.sub.1]-1)[t.sup.d] [dela.sub.2.t.sup.i] - [dela.sub.3.t.sup.e +...]

Employers' real wage costs are average earnings plus employers' contributions deflated by producer prices, or in log terms (w-[p.sup.p]+ [t.sup.e]), where pp is the log of producer prices. if we assume for expositional purposes that consumer prices are simply producer prices plus a mark-up for indirect taxes (ignoring, for example, import prices) so that [p.sup.c] = [p.sup.p] + [t.sup.i] then substituting from (1) gives

(3) Employers' real wage costs (log) = w - pp + t'

= W - (p.sup.c] - [t.sup.i]) + [t.sup.e]

= [dela.sub.1.t.sup.d] + (1 - [dela.sub.2])t.sup.i + (1-[dela.sub.3])[t.sup.e] +...

Although in full model simulation other variables such as unemployment impinge on wages we can use equations (2) and (3) to derive the direct effect of a cut in any of the tax rates on the workers' real consumption wage and employers' real wage costs. For example, considering a fall in income taxes M, if none of this cut is absorbed in lower nominal wages al = 0) then the tax cut is completely reflected in a rise in the real consumption wage and then operates as a shock to demand. Similar results arise for [dela.sub.2] =1 or [dela.sub.3] = 1, for cuts in [t.sup.i] and [t.sup.e]. At the other extreme, if [dela.sub.] = 1 then employers' real wage costs fall by the full extent of the income tax cut, hence lowering the NAIRU. Again identical results occur with [dela.sub.2] = 0 or [dela.sub.3] = 0 for a fall in [t.sup.i] and [t.sup.e] In practice the only non-zero values of the x-coefficients in the quarterly models' wage equations are as follows: for HMT [dela.sub.1] = 0.6; for BE [dela.sub.1] = 0.2 (on average allowing for interaction between wage sectors); and for NIESR [dela.sub.1] = 0.47, 12 = 0.53 and [dela.sub.3] = 0.53. A notable feature of the NIESR coefficients is the imposition of the restriction (1 - [dela.sub.1]) = [dela.sub.2] = [dela.sub.3], which implies that the long-run effects of equivalent changes in the tax rates [t.sup.d, t.sup.e] and [t.sup.e] are identical. However it is also worth noting that NIESR themselves concede that the 'estimated coefficients on individual tax effects were often not robust'. Income tax cut Differences in the size of the retention ratio effect, [dela.sub.1] equation (1), continue to be an important explanation of differences in this simulation. The retention ratio is the ratio of the net wage to the gross wage where the former is defined as gross earnings less income tax and employees' national insurance contributions. This variable is an important determinant of wages if it is assumed that workers bargain for a target real consumption wage.

Given that there is no long-run retention ratio effect in either the LBS or OEF models (although in the latter case there is some temporary short-run effect) much of the initial stimulus to GDP therefore comes from higher consumption as a result of increased disposable income. In both cases this is also reinforced by similar mechanisms to those described for the government expenditure simulation. The absence of any long-run retention ratio effect means that by years 4-5 the inflation-unemployment tradeoff in the LBS and OEF simulations is similar to that under the government expenditure simulation, perhaps not a surprising result given that both policy changes can simply be regarded as a shock to aggregate demand, although increased government expenditure delivers a much more rapid fall in unemployment. In the OEF simulation a notable feature of the short-term results is that there is a complete absence of any rise in prices over the first two years, a result which is explained by a powerful short-run retention ratio effect, although as can be seen by the subsequent response of real wages, this seemingly has little influence on the longer-run simulation results.

In the BE simulation there is a small long-run retention ratio effect, [dela.sub.1] = 0.2, and this entails a fall in unemployment (and rise in GDP) similar to that in the government expenditure simulation by year five, but with an increase in the consumer price level of only two-thirds that in the government expenditure case. The influence of a much larger retention ratio effect ([dela.sub.1] = 0.6) can be seen more clearly in the HMT simulation, where there is a more substantial fall in employers' real wage costs and hence greater downward pressure on consumer prices. By the fifth year of the simulation the rise in the consumer price level is only about one-fifth of the rise following the government expenditure increase, but unemployment has fallen by slightly more than in the previous simulation.

The influence of the retention ratio effect is less immediately discernible in the NIESR simulation because the domestic price level still rises in response to the step fall in the nominal exchange rate. Nevertheless employers' real wage costs do fall and hence put downward pressure on inflation thus allowing a long-run fall in the NAIRU. By the fifth year of the NIESR simulation the model appears to be converging on this new NAIRU with an overall fall in unemployment of about 40 thousand, similar to that achieved in the government expenditure simulation. There are some important differences between the two simulations, however. For example, because the present result does not rely on the appreciation of the real exchange rate that occurs in the government expenditure simulation, there is an improvement rather than a decline in manufacturing output. A cut in VAT. The direct effect of a cut in VAT by one percentage point is to reduce the price level by less than 1 per cent in the NIESR, BE, HMT and OEF models (reflecting the fact that VAT does not cover all goods and services). In the case of LBS the single-equation response of the consumer prices equation implies that a change in VAT only has a temporary effect on the price level. The cause of this counter-intuitive property can be traced to the functional form in which VAT enters this equation (also highlighted in our previous review of model properties). A re-estimate of the LBS equation using the NIESR measure of indirect taxes in a more suitable functional form allows the imposition of a restriction which implies that changes in indirect tax rates have, allowing for coverage, a one-for-one effect on consumer prices. This feature of the original LBS equation makes it difficult to interpret the VAT simulation, although it would seem that the effects of a cut in VAT are very similar to those of a cut in income tax.

The absence of any indirect tax effects in the HMT, BE and OEF wage equations ([dela.sub.2] = 0) means that the direct effect of the cut in VAT is largely reflected in a fall in employers' real wage costs. This reduction is then passed on in lower prices which in turn leads to an appreciation of the exchange rate and further downward pressure on prices and so on, so that beyond the first year the fall in the price level is significantly more than the direct effect of the cut in VAT. This downward spiral in prices is counteracted in the BE simulation by downward pressure on the real exchange rate resulting from the deterioration in the current balance and the increase in the PSBR; and in the HMT and OEF simulations by the lack of homogeneity in the exchange-rate equation. A further limiting factor which is common to these three models is the upward pressure on real wages from both labour market and goods market sources. These counteracting inflationary pressures eventually retard the fall in the price level in the HMT and OEF simulations, although the fall in the price level in the BE simulation shows no sign of slowing.

The effect of a cut in VAT on the NIESR model is very similar to a cut in the basic rate of income tax particularly in the long run, a result which is consistent with the above analysis, the discrepancy being due to differences in the implied change in average tax rates in the simulations. In the short run the VAT cut has some advantages over the income tax cut from its more immediate deflationary impact on the price level, so that the initial increases in the price level are slightly less in the VAT simulation. A cut in employers' NICs. Given that the coefficients on employers' contributions in the model wage equations are identical to those on indirect taxes in the HMT, NIESR, BE and OEF models (i.e. ot, = oc,) it is not surprising that a cut in employers' NiCs has very similar effects to a cut in VAT in all these models, although it should be borne in mind that the magnitude of the two tax changes are not identical. The main difference would seem to be that in all four mode(s the cut in VAT has a much more immediate impact on the price level, whereas additional lags follow the cut in employers' NICs before lower labour costs are passed on in lower wholesale and consumer prices. An additional reason why a cut in employers' NICs has a less immediate impact is that both the HMT and BE wage equations imply that the cut in NiCs leads to a temporary increase in wage demands, as workers perceive that firms can afford higher pay increases, although there is no such permanent effect (i.e. although [dela.sub.3] = 0, in the short run there is a positive coefficient).

In the LBS simulation the effect which the cut in employers' NICs has on lowering wage costs is substantially outweighed by the inflationary effect of the fall in the exchange rate, so that consumer prices rise (as is the case for NIESR). However it is puzzling that the inflationary effects in this simulation are at least as great, when normalised on either GDP or unemployment, as for simulations of an income tax cut or an increase in government expenditure on this model. Problems with the specification of the wholesale price equation, namely sluggish adjustment, noted above, and the comparatively low weight which wage costs have in determining prices compared to other costs, contribute to this result. In addition, for no obvious reason, employers' NICs are excluded altogether from the definition of the wage cost variable which appears in the consumer price equation, even though they are included in the wage cost variable which appears in the wholesale price equation.

Cut in short-term nominal interest rates In this exercise we consider the effects of reducing short-term interest rates by 1 percentage point. Term structure equations in the models relating long rates to short rates are assumed to continue to operate. In the LPL model interest rates are jointly determined with the exchange rate to maintain uncovered interest-rate parity and hence this experiment is not performed for this model. The exchange rate in the NIESR model can be regarded as an uncovered interest-rate parity condition with the inclusion of a term in the ratio of the net acquisition of overseas assets to GDP, reflecting a risk premium. This produces the result that the initial fall in the exchange rate for a permanent change in interest rates depends on the length of the simulation period and hence can produce an excessively large initial reaction. In our simulation the shock lasts for the five years of the reporting period but it is assumed that expectations are formed on the basis that the shock does not persist thereafter and hence is regarded as temporary.

In examining the transmission mechanism in the quarterly models through which interest rates operate on consumption, investment and the exchange rate, we distinguish between direct and indirect effects. The direct effects are the single-equation partial responses, and the full simulation results (table A5) are then the sum of these direct effects plus indirect effects which occur from feedback through all the model equations. The direct effects on consumption range from 0.32 per cent for LBS to 0.65 per cent for BE, although in the latter case there are very long lags, the median lag for the effect of interest rates on non-durable consumption being about four years. There is general agreement across the models that, proportionately, the largest direct effects are on private housing investment, ranging from 1 per cent for OEF to 3.3 per cent for BE. However while all of the models, with the exception of LBS, have significant direct effects on manufacturing investment, there is much greater disagreement on the relative effects on non-manufacturing investment. In those models which make little attempt to disaggregate the non-manufacturing sector (LBS, HMT and OEF) there are no long-run direct interest rate effects, whereas those models which do disaggregate the non-manufacturing sector (NIESR and BE) find significant direct interest-rate effects. It is the latter models where, not surprisingly given the importance of the non-manufacturing sector in determining total investment, the overall direct effects of interest rates are clearly larger on investment than on consumption. This result is particularly true for the BE model where a large interest-rate effect on distribution and other services' contributes to the result that every 1 percentage point fall in interest rates increases total fixed investment by about 1 per cent.

There are also substantial differences in the response of the exchange rate to interest rates. The direct effect of every 1 percentage point cut in interest rates is to lower the real exchange rate by 1.3, 0.6 and 0.5 per cent according to the HMT, BE and OEF models respectively. In the NIESR model a fall in interest rates of 1 percentage point which is expected to last for 5 years leads to an approximate fall in the nominal exchange rate of about 5 per cent, with the exchange rate subsequently appreciating by about 1 per cent per annum. Given that the exchange rate is determined by the complete interaction of the LBS financial sector it is not possible to calculate the direct effect of a cut in interest rates for this model. As previously argued this can be a serious limitation to understanding full model simulation properties, a point which is particularly relevant in the present case where beyond the first year the cut in interest rates produces the counter-intuitive result of an appreciation in both the nominal and the real exchange rate. The rise in the real exchange rate, together with comparatively weak direct effects on consumption and investment, explains why the GDP effects are so small in the LBS simulation.

The NIESR simulation results are also dominated by the response of the exchange rate. Thus in the first year of the simulation the large fall in the nominal exchange rate improves competitiveness and leads to the largest GDP response of any of the models. However, as the nominal exchange rate subsequently rises and prices increase, the improvement in competitiveness and hence the GDP response are dampened.

In the HMT simulation the overall stimulus to GDP is very much larger than that suggested by the direct effects on consumption and investment. This is partly explained by the fall in the real exchange rate, although the absence of price homogeneity in the exchange-rate equation means that this fall is not as large as the direct effect. in addition there is a further stimulus to consumption from the effect which interest rates have on improving real wealth, and by the fourth year the rise in consumption is nearly 9 per cent. The total effects on fixed investment are also much larger than the direct effects, particularly due to a very large accelerator response from non-manufacturing investment: beyond the third year the total rise in fixed investment is more than 3 per cent. These accelerator effects on fixed investment operate even more strongly in the BE simulation so that in combination with large direct interest-rate effects they produce a rise in fixed investment of about 4 per cent by year 5, which is nearly three times the percentage increase in consumption. Given the absence from the OEF model of any large direct interest-rate effects on either consumption, investment or the exchange rate, the overall GDP response in this simulation is significantly less than for either HMT or BE, with a more balanced increase (in percentage terms) in the expenditure components.

4. Conclusion

There have been important developments in the supply side of the main large-scale macro models, and these have changed their overall properties, as revealed in standard simulation experiments. In particular, expansionary demand policies now tend to be associated with smaller increases in domestic output in the longer run than in earlier vintages of these models (see, for example, Wallis et aL, 1986). The principal cause of this change is the greater degree of inflationary pressure generated by a demand expansion, and the subsequent feedback effects of higher inflation in reducing demand. In the short run, however, the quarterly models appear very Keynesian. For example, although the NIESR model can no longer be classified as Keynesian in respect of its simulation responses at the end of a five-year period, it exhibits these properties in the short term because wages and prices adjust relatively slowly despite rational expectations in wage and price setting.

Greater emphasis on the supply side and the inflation mechanism in the quarterly models is a response to the increased focus on these issues in policy making. It can also be seen as a response to the challenge of the LPL model, first used for forecasting in 1980, and placing great emphasis on the supply side of the economy. The overall specification of the LPL model relies heavily on predefined economic theory, thus many of its properties are known a priori, and these have remained largely unchanged since the model's inception. It is an equilibrium model which assumes market clearing in all sectors of the economy and corresponds to the new classical paradigm. it remains an outlier, however, many of its properties being at variance with those of the quarterly models which, despite giving greater emphasis to the supply side, reflect a stronger reliance on empirical evidence, and in particular the view that many markets do not correspond to the perfectly competitive paradigm. Simple monetarist models of inflation, prevalent in the early 1980s, have been proved inadequate, and the quarterly models now attempt to provide fully elaborated representations of the inflation mechanism.

Differences in the long-run properties of the quarterly models can be largely explained by differences in the key wage-price-exchange-rate interactions. in turn these differences can often be resolved by examining the economic and statistical credentials of these structural relationships. However they also arise because different groups attach different relative weights to economic theory and empirical evidence in the detailed specification of the model, and two examples stand out in this paper. The first is the issue of homogeneity in the wage, price and exchange-rate equations. Some models, for example NIESR and HMT, place a high weight on the theoretical requirement of homogeneity, whereas the OEF model relies more on statistical estimates of the relevant parameters and does not impose the corresponding homogeneity restrictions. This difference of emphasis has important implications for overall model properties, and specifically for the nature of the long-run response in the OEF model.

The second example concerns the role of taxes. Our analysis in Section 3 shows that estimates of the response of the economy to a change in tax rates, both between models and between different types of taxes, can be explained largely by inspection of the parameters of the wage relationships. These tax coefficients tend to be poorly determined in empirical estimation, however, partly because the aggregate tax variables do not adequately reflect changes in the particular taxes which influence decisions at the micro level. In response to this uncertainty some modellers, specifically NIESR, choose to impose the theoretical prior that the long-run effect of different tax changes on wages, and hence the NAIRU, should be equal, whereas others rely on statistical evidence, which often fails to indicate that a significant non-zero effect is present. Again this difference has important implications for model-based policy analysis, and its resolution may require a more disaggregated treatment.

Of the quarterly models considered in this paper only that of NIESR can be said to possess a clearly defined long run, although all the models have incorporated more supply-side factors in their structure. Further research is required before it can be determined whether the current divergencies from the NIESR framework observed in the other models represent genuine differences in approach or differences in detailed specification that can be empirically resolved. Despite differences in the treatment of the supply side the models nevertheless represent a useful common framework within which to interpret the inflationary processes in the economy and, more generally, through ready-reckoners, to provide a quantitative assessment of policy options.

REFERENCES

Courakis, A.S. (1988), Modelling portfolio selection', Economic Journal, 98, pp. 619-641. Fisher, P.G. and Turner, D.S. (1990), The exchange rate, forward expectations and the properties of macroeconometric 'models', Discussion Paper No.23, ESRC Macroeconomic Modelling Bureau. Fisher, P.G., Tanna, S.K., Turner, D.S., Wallis, K. F. and Whitley, J.D. (1989), Comparative properties of models of the UK 'economy', National Institute Economic Review, No. 129, pp. 69-87. Fisher, P.G., Tanna, S. K., Turner, D. S., Wallis, K. F. and Whitley, J. D. (1 990), 'Econometric evaluation of the exchange rate

In models of the UK economy', Economic Journal, forthcoming. Gregg, P. (1989), 'Out for the count-a new approach to modelling claimant unemployment', Discussion Paper No. 167,

National Institute of Economic and Social Research. Joyce, M. and Wren-Lewis, S. (1989), 'Does the NAIRU matter?', Discussion Paper No.1 54, National Institute of Economic

and Social Research. Layard, P.R.G. and Nickell, S.J. (1985).'The causes of British unemployment', National Institute Economic Review, No.1 1 1,

pp. 62-85. Melliss, C.L. (1988), 'HM Treasury macroeconomic model 1986', Economic Modelling, 5, pp. 237-259. Melliss, C., Meen, G., Pain, N. and Whittaker, R. 1989),'The newtreasury model project', GES Working Paper No.106. Turner, D.S., Wallis, K.F. and Whitley, J.D. (1989), Differences in the properties of large-scale macroeconometric models:

the role of labour market specifications', Journal of Applied Econometrics, 4, pp. 31 7-344. Wallis, K.F. (ed.), Andrews, M.J., Fisher, P.G., Long bottom, J.A. and Whitley, J.D. (1986), Models of the UK Economy: A

Third Review by the ESRC Macroeconomic Modelling Bureau, Oxford: Oxford University Press. Wren-Lewis, S. (1989), The macroeconomics of supply', Discussion Paper No.160, National institute of Economic and

Social Research.

Printer friendly Cite/link Email Feedback | |

Title Annotation: | United Kingdom |
---|---|

Author: | Fisher, P.G.; Turner, D.S.; Wallis, K.F.; Whitley, J.D. |

Publication: | National Institute Economic Review |

Date: | Aug 1, 1990 |

Words: | 8620 |

Previous Article: | The growth of output and productivity in the UK: the 1980s as a phase of the post-war period. |

Next Article: | Summary and appraisal. |

Topics: |