Printer Friendly

Comparative properties of models of the UK economy.


This paper describes the properties of five major macroeconometric models of the UK economy, through analysis of six standard policy simulations. The simulations are conducted, as far as possible, in a consistent manner across the models. The models were deposited with the ESRC Macroeconomic Modelling in late 1988 and this paper follows comparative analysis of previous model vintages in this Review and in earlier Bureau publications. The paper highlights differences between the models that emerge from the treatment of exchange rates, imports, consumption and price adjustment. Recent changes to the models are shown to have increased simulation differences between them.

1. Introduction This paper describes the properties of five major macroeconometric models of the UK economy, through analysis of a number of standard policy simulations. The models are those of the London Business School (LBS), the National Institute of Economic and Social Research (NIESR), Her Majesty's Treasury (HMT), the Bank of England (BE) and the Liverpool University Research Group in Macroeconomics (LPL), as deposited with the ESRC Macroeconomic Modelling Bureau in late 1988. The simulations are conducted, as far as possible, in a consistent manner across the models.

Four of the five models considered are large quarterly models (LBS, NIESR, HMT, BE) and the remaining model is a small annual model (LPL). The latter (Minford et al., 1984) has less than 20 behavioural equations, and it represents the new classical/rational expectations approach to macroeconomic policy analysis. This model has remained largely unchanged over the last five years and hence there is little discussion of its simulation properties in this paper. All of the quarterly models originate from the Keynesian income-expenditure approach, although there has been an increasing tendency over recent years to incorporate aggregate supply considerations. In particular the latest versions of the LBS and NIESR models include important changes to their structure in this respect.

The major change to the LBS model (Dinenis et al., 1989) is the inclusion of a consistently estimated set of equations determining the demand for and price of domestic output. Distinctions are made both between the manufacturing and non-manufacturing sector and between domestic and foreign demand for home producers' output. The share of home producers' output in total domestic demand is a function of relative (domestic to import) prices and a time trend, although the equation can be reparameterised in a more conventional form which determines the share of imports in domestic demand in terms of the same explanatory variables. Price adjustment now plays a greater role: there is upward pressure on prices if output increases relative to the capital stock or if input prices increase. Dinenis et al. (1989) show how the long-run solution to the price equations can be inverted in order to obtain implicit long-run aggregate supply functions where domestic output is dependent on the capital stock and the relative price of inputs to output. One area of the LBS model which remains unchanged is the disaggregated model of the financial sector, which emphasises the role of forward expectations and which, inter alia, determines the behaviour of the exchange rate.

The NIESR model (Wren-Lewis, 1988) has also changed substantially since our previous review (Fisher et al., 1988). Of particular note is the incorporation of a putty/clay vintage production function for the manufacturing sector which relates the level of productivity to investment, and determines capacity utilisation. The latter variable has an important role in the investment and import equations and also in the price adjustment process, where any increase in manufacturing output which is not matched by increased capacity leads to inflationary pressure from a higher profit mark-up. The model retains forward expectations in several areas, notably that of exchange-rate determination.

The HMT model (Melliss et al., 1989) remains the largest of the quarterly models (some 400 endogenous variables) as a result of its detailed treatment of the public sector, although the latest public release is greatly simplified and reduced in size in relation to its predecessor. The BE model (Harnett and Patterson, 1989) contains a relatively detailed treatment of the monetary sector and has also been considerably slimmed down compared with recent versions. Both of these models contain important supply-side elements with, for example, wages and prices adjusting in response to changes in demand, and factor Comparisons across the models are based on six standard policy simulations, as follows:

(i) An increase of 2 billion [pounds] (1989 prices) per

annum in government current expenditure on

goods and services (representing about 2 per

cent of real current government expenditure, or

1/2 per cent of total GDP) (ii) A reduction in the standard rate of income tax of

1 percentage point (iii) A 1 percentage point reduction in the rate of

employers' national insurance contributions

(NICs) (iv) A 1 percentage point reduction in the standard

rate of VAT

(v) A reduction of 1 percentage point in nominal

short-term interest rates (vi) A reduction of 10 per cent in the rate of

unemployment benefit. All changes are assumed to be unanticipated and permanent (except for the interest-rate shock in the NIESR model). Experience suggests that the simulation responses can be used as ready-reckoners to calculate the response of target variables to a range of shocks to the policy instruments, singly or in combination. The simulation experiments are conducted around a base run which, in the case of the `non-official' models, corresponds to the model proprietor's published forecast and hence incorporates all the assumptions and adjustments made by the forecasters. For the `official' models, a simulation base is either provided by the model proprietor (BE) or constructed by ourselves (HMT). General results on the major macroeconomic indicators over a five-year horizon, the longest period common to all the models, are presented in tables 1-6. Our discussion focuses in particular on output, unemployment and inflation. A technical note giving details of the experimental design is available from the authors on request.

Simulation exercises with macroeconomic models require some view to be taken of the accompanying stance of monetary and fiscal policy. In general, any shocks may lead directly or indirectly to changes in the public sector borrowing requirement (PSBR), and different results may be obtained according to the way in which this is financed. The standard possibilities are that the change in the PSBR is accommodated by additional money creation (money finance) or financed by the sale of government debt to the non-bank private sector (bond finance). On a practical level, many of the models contain an implicit overall policy stance as a kind of base setting, and sometimes this is not easily modified. The policy setting most readily available for the quarterly models is that of constant nominal short-term interest rates, which approximates money finance. This requires the suppression of the relationships in the LBS, NIESR and BE models whereby interest rates are determined by a policy reaction function. In the case of LPL the required addition to the money stock is not easily applied, and here the simulations assume that any change in the government deficit is financed by an equi-proportionate change in money and bonds, referred to as `balanced finance'.

The article proceeds as follows. Section 2 examines in detail the government expenditure simulation under an assumption of fixed nominal interest rates (balanced finance for LPL), dealing in turn with the response of GDP, employment and unemployment, prices and the PSBR. We emphasise this simulation since many important features of a particular model are common to other simulations on the same model. Discussion of the remaining simulations in Section 3 then concentrates on the additional features that emerge. Section 4 contains concluding comments. Our aim is not simply to report the results of standard simulation exercises but also to identify and explain the sources of major differences across the models. Earlier exercises, for example (Turner et al., 1988), have proved that it is possible to reconcile conflicting simulation results by analysis of particular features of model equations.

2. Government expenditures simulation In this section we examine the effects of an increase in government current expenditure under an assumption of fixed nominal interest rates (LBS, NIESR, HMT and BE) or balanced finance (LPL). The increase is allocated proportionately between procurement and employment in those models that make this distinction (LBS, HMT) and between central and local government where this additional disaggregation occurs (HMT).


The GDP multipliers (defined as the ratio of the absolute increase in GDP to the increase in government expenditure) are shown in table 7. They reveal three broad responses in the models; those where the multiplier is consistently small (LPL), those where it remains around unity (HMT, BE), and those where the multiplier varies quite markedly over the simulation period (LBS, NIESR). In the first year the increase in GDP is greatest for NIESR: by the second year of the simulation both NIESR and LBS give similar multipliers but thereafter there is a marked divergence between the GDP effects in these two models, with a sharp downturn in the NIESR multiplier (which falls to only 0.4 by the fifth year) whereas the LBS multiplier rises steadily (to a value of 2.2). The GDP effects in the HMT and BE simulations exhibit much greater stability with multipliers which are close to unity.

The impact multiplier for LPL is actually negative and beyond the second year there is only a negligible effect. In our previous review we noted that the pattern of response of the GDP multiplier in the LPL simulation was dominated by the behaviour of durables expenditure (stockbuilding, investment and consumer durables). A key feature of this equation is the presence of a wealth term in place of current income and the effect of anticipated inflation on this variable results in a fall in the private sector stock of durables of more than three times the increase in government expenditure in the first year of the simulation, with a rebound in GDP in the second year, when most of the stock adjustment is complete. This effect persists in the current version of the model despite the finding that the coefficients of this particular equation are not consistent with the data (Wallis et al., 1987, Chapter 5.4).

Exchange rate

Among the quarterly models the behaviour of the exchange rate is important in explaining the differences in GDP response. In simulations where the real exchange rate is held constant the GDP multipliers average unity. Research by the Bureau (Fisher et al., 1989) has shown that it is possible to derive a preferred exchange-rate equation from those included in the models, and that this resolves much of the differences in model simulation responses.

In the standard model simulation the two quarterly models with forward consistent expectations (LBS and NIESR) exhibit an initial jump in the nominal exchange rate in the face of a demand shock but the subsequent behaviour of the real exchange rate differs markedly. In the case of LBS a sluggish response of domestic prices results in the real exchange rate remaining lower by between 1 1/2-2 1/2 per cent over the simulation period, thus stimulating private sector output through net trade. Conversely, in the NIESR simulation the gains to competitiveness from the once-for-all step fall in the nominal exchange rate of 3 1/2 per cent are steadily eroded by inflation. Thus beyond the third year there is no improvement in the real exchange rate, and the absence of this stimulus combined with the crowding-out effects of higher inflation on consumers' expenditure causes the sharp downturn in GDP, with almost complete crowding out of GDP. These NIESR model results are in marked contrast to those discussed in Fisher et al. (1988) where the fall in the exchange rate produced a more permanent improvement in competitiveness which substantially enhanced the GDP response over the entire simulation. This change can be partly attributed to the introduction of a new exchange-rate equation in which any unanticipated policy change causes an initial jump in the nominal exchange rate sufficient to eliminate any difference in the current account at the end of the simulation.

The BE model includes a new exchange-rate equation which produces a fall in the real exchange rate in response to imbalances in non-oil trade, with every 1 billion [pounds] deterioration leading to a fall in the real exchange rate of about 0.7 per cent. In the government expenditure simulation this produces a gradual fall in the real exchange rate ranging from 0.2 per cent in the first year to 0.4 per cent in the fifth year. The stimulus to private sector output from this improvement in competitiveness accounts for about half the (small) increase in the government expenditure multiplier over the five years.

The exchange rate in the HMT model is insensitive to changes in the current account. The nominal exchange rate falls in line with domestic money aggregates but, since these do not rise as rapidly as domestic prices, the real exchange rate rises; this explains why the GDP multiplier remains fairly stable in the HMT government expenditure simulation.

An understanding of differences in GDP response between the models is helped by examination of the different expenditure components of GDP. In the first year of the simulation the GDP response is dominated by the autonomous government spending increase but changes in the endogenous expenditure components do lead to short-run differences in output. By the fifth year of the simulation some of these responses, considered next, are highly significant in augmenting or offsetting the exogenous expenditure shock, and hence in explaining large differences in output response across the models.

Exports and imports

The contribution to GDP from net trade follows to a large extent from the real exchange rate discussed above. Thus the largest increase in exports is initially in the NIESR and LBS simulations, where exports account for about 0.3 of the government expenditure multiplier (gross of any import content) rising to just under 0.5 for the remainder of the simulation. However the gains in competitiveness are gradually eroded in the NIESR simulation so that by the fifth year there is no contribution from increased exports. In the BE simulation the more modest fall in the real exchange rate brings about a smaller increase in exports which account for up to 0.15 of the government expenditure multiplier. In the HMT simulation, where there is a slight rise in the real exchange rate, the change in exports is negligible.

The multiplier response in the LBS and NIESR simulations is further boosted by a comparatively weak response from imports in relation to the increase in total final expenditure. In comparison the growth in imports is relatively much higher than that of total final expenditure for both HMT and BE. These model differences are due not only to exchange-rate behaviour but also to different specifications for manufacturing imports. Import behaviour is an important route for supply-side influences in the models. In the NIESR model the share of manufacturing imports in demand is determined by relative prices, capacity utilisation and a specialisation variable, so that the share of imports may vary cyclically but, for constant capacity utilisation, the elasticity of manufacturing imports with respect to the demand for manufactures is imposed at unity. The HMT model uses a standard log-linear relationship between imports, demand, relative prices, relative capacity and specialisation, with a demand elasticity of 1.5. This difference in demand elasticity is partly accounted for by the stronger effect of trend specialisation variables (which remain constant in simulations) in the NIESR model. In the LBS and BE models it is the share of gross manufacturing supply to the domestic market which is made a function of domestic demand, with imports then calculated as the difference between these aggregates. Although in principle this approach is no different from that of NIESR practical considerations result in different estimates of the input propensity of domestic expenditure. First, the demand variable in the LBS model is the simple aggregate of domestic expenditure (excluding government current expenditure on wages and salaries) whereas BE, in common with NIESR and HMT weight together the various categories of total final expenditure according to their manufacturing content. Second, the transition from gross output to net output in the LBS model produces a much higher average import content of exports and hence a much lower import content of domestic expenditure than do either HMT or BE. These differences partly explain the comparatively low import propensity in the LBS simulations.

Consumers' expenditure

This plays a central role in the multiplier process following any policy change, and because it accounts for such a large proportion of total final expenditure, differences in the specification of consumption equations can be particularly important in explaining differences in model properties. In comparing the effects on consumers' expenditure across the models NIESR is an obvious outlier, with virtually no change in consumers' expenditure in the first year and a fall in consumption thereafter, despite increases in real disposable income. In contrast the growth in consumption is positive in all the other models, rising from about a third (in the case of BE) or just under a half (LBS and HMT) of the percentage change in real personal disposable income in the first year to about 0.7 (HMT and BE) or 0.8 (LBS) by the fifth year. Across all of the models the single-equation response implies a long-run elasticity of total consumption with respect to real income of approximately unity and the most important explanation of why this response is not observed in full-model simulations is the effect of inflation. Inflation can influence consumer behaviour through a variety of channels. In all of the models, consumers' expenditure is dependent on the real value of financial wealth or real liquid assets. However in most simulations on the LBS, HMT and BE models these wealth variables maintain their real value despite increased prices. However in the NIESR simulations about 70-80 per cent of any increase in prices is reflected in a fall in the value of real liquid assets and about 40-50 per cent in a fall in net personal financial wealth. The relatively greater increase in prices in this model produces a wealth effect which is sufficient to outweigh the effect of increased real disposable income. Inflation can affect consumption directly or via real interest rates, although in the LBS model these effects are very weak, whereas in the HMT and BE models, although they are much stronger, the inflation effects to a large extent cancel each other out. The most important crowding-out effects on consumption in the HMT and BE models arise because the income variable used in their consumption equations is defined to include inflationary losses on personal sector liquid assets and the rise in inflation-adjusted income measure averages around 40 per cent of the increase in real disposable income.

Borrowing also plays a role in the NIESR equation for (total) consumption, with all changes in real consumer credit being reflected one-for-one in higher consumption. However in most of the simulations we report here, with the possible exception of changes to interest rates, there is little influence from this credit variable on consumption.

Investment and stockbuilding

A second important route mechanism by which the fall in the exchange rate is translated into higher GDP in the NIESR simulation is through accelerator effects as stockbuilding and investment respond to higher private sector output. By the second year of the simulation the increases in investment and stockbuilding together contribute 0.8 to the GDP multiplier (gross of any import content). Moreover a sharp turnround in stockbuilding (so that there is de-stocking) accounts for much of the turnround in GDP beyond the second year. In most of the models accelerator-type relationships for stockbuilding tend to magnify any fluctuations in private sector output, although the speed of adjustment is greatest in the NIESR model where stockbuilding reacts to consistent forward expectations of output. The largest increase in fixed investment in the NIESR simulation is in distribution and business services, where the short-run response of investment relative to output is substantially greater than the long-run response, constrained to unity. This also applies to non-manufacturing investment in the HMT model and to a lesser extent distribution and business services investment in the BE model. This dynamic response of non-manufacturing investment is not a feature of LBS simulations where there is a short-term fall, although the non-manufacturing investment response to output is approximately unity in the long run.

Employment and unemployment

The increase in government sector employment in the government expenditure simulation is about 100-115 thousand across all the quarterly models and accounts for most of the increase in employment in the early years (although in the BE model, unusually, it takes a year before government employment fully responds to the change in government expenditure). There is little disaggregation of private sector employment beyond manufacturing and nonmanufacturing, with sectoral output being the main determinant of private sector employment. In the case of manufacturing and non-manufacturing employment in the HMT and BE models and non-manufacturing employment in the NIESR model a long-run unit elasticity is imposed between employment and output. In the LBS equations for manufacturing and non-manufacturing employment the capital stock enters as an explanatory variable and constant returns to scale are imposed. Elasticities with respect to sectoral output are significantly greater than unity, counterbalanced by negative elasticities with respect to the capital stock. Since in most simulations the change in output is much greater and more immediate than the change in capital stock, this tends to produce comparatively large changes in employment in the case of the non-manufacturing sector. Thus beyond the second year in the government expenditure simulation the percentage increase in non-manufacturing employment is 1-1.8 times the percentage change in non-manufacturing output. The effect is less marked for manufacturing due to a comparatively sluggish response from manufacturing employment (the equation dynamics imply median lags with respect to output and the capital stock of about five years). The desired level of manufacturing employment in the NIESR model is determined by the vintage production system, with actual manufacturing employment then adjusting gradually to the desired level. The changes in manufacturing employment which this produces do not appear out of line with the results of the other models, manufacturing employment broadly following manufacturing output, but with a substantial lag. Whereas the LBS model is an outlier with respect to the determination of employment, the more striking unemployment effects occur in the BE model.

In particular the BE results imply that any fall in unemployment is a very low proportion of the increase in employment in comparison with the other models. On average, the ratio of the fall in unemployment to the increase in employment is about 60 per cent as the increased availability of jobs may encourage more people to enter the working population. In the BE model the fall in unemployment is only 13 per cent of the number of jobs created in the first year and 25 per cent in the fifth year. This result is caused by an unemployment equation which has the property that while manufacturing jobs reduce unemployment on practically a 1:1 basis, unemployment falls by only 16 per cent of any increase in non-manufacturing (including government sector) employment. Turner et al. (1989), however, find that this is not a reliable estimate.

In addition to determining the change in total unemployment the BE, NIESR and LBS models also make a distinction between short-term and long-term unemployment, which is important in the determination of wages. However the changes in the duration structure of unemployment in the BE simulations appear to be implausible, since they show falling total unemployment together with rising short-term unemployment. This result is caused by unsatisfactory features of the equation determining short-term unemployment, and Turner and Whitley (1989) give an alternative specification of this equation (along the lines adopted by NIESR) which has more acceptable properties.

Wage and price effects

Wage and price responses are another important source of potential supply-side influence. In the LBS, HMT and BE simulations the reduction in unemployment stabilises, without any obvious sign of accelerating inflation. After 3-4 years the inflationary cost of reducing unemployment by 100 thousand is about 0.5-0.6 per cent per annum on the HMT and BE models and about 0.35 per cent per annum on the LBS model. However GDP--inflation trade-offs show that the HMT model is most inflationary, followed by LBS and then BE. As neither GDP nor inflation settles down over the later years of the NIESR simulation the calculation of any trade-off for this model is not very meaningful; nevertheless the total increase in the price level by the fifth year of the NIESR simulation is over twice that obtained in any other quarterly model.

In the NIESR model demand expansion induces a step fall in the exchange rate and, given that static homogeneity exists in the wage and price system of this model, the additional increase in import prices is fully passed through to domestic wages and prices. Adjustment is quite rapid; this reflects the single-equation response for both consumer and wholesale prices which show some tendency to overshoot their long-run effects. Demand--pull effects originating directly from the labour market are relatively weak, partly since only short-term unemployment influences average earnings, but further inflationary pressure emerges from the effect of higher demand in the goods market. Increases in capacity utilisation (derived explicitly from the vintage production approach) raise the price mark-up on costs permanently, leading to an increase in the price level of around 0.6 per cent.

The LBS simulation exhibits much weaker inflationary tendencies from the demand expansion. The exchange rate declines initially by a similar extent to that of NIESR, but the absence of static homogeneity in the wage and price system of this model prevents a full pass-through of the effects of higher import prices. The most important source of non-homogeneity is the equation for manufacturing input prices, where a 1 per cent increase in import prices of fuels and non-oil commodities (the sole determinants of import prices) leads only to an increase of 1/2 per cent in manufacturing input prices. Absence of static homogeneity arises from the specification of the equation in first differences, but our re-estimation demonstrates that levels terms can be included and that the restriction of static homogeneity can then be imposed and accepted. In the LBS model each 1 per cent fall in total unemployment leads to an increase in real consumption earnings of about 0.06 per cent and, although the effect on unit labour costs is mitigated by an increase in manufacturing productivity, there is further upward pressure on unit labour costs as productivity in non-manufacturing falls. Pressure of demand in the goods market is incorporated in the equation for wholesale prices through a term in the ratio of output to the capital stock. As noted above, output responds more rapidly than does capital stock and hence the implied increase in capacity utilisation generates greater inflationary pressure in the LBS model. However, the lack of a direct relationship from wholesale prices to consumer prices (the latter being determined as a function of import costs, labour costs, and indirect taxes), together with the sluggish adjustment of wholesale prices to changes in capacity utilisation, results in this source of inflationary pressure being considerably dampened.

The limited change in exchange rates in the HMT and BE models implies little impact on the price level from this source. The single-equation effect of unemployment on wages originating in the labour market for HMT are of similar magnitude to those of LBS, although in full-model simulations this leads to much greater pressure on prices than does that of LBS. Labour market pressures are comparatively weak for BE. Not only does total unemployment fall modestly in the BE simulation but, as for NIESR, it is only short-term unemployment which influences earnings and, counter-intuitively, this actually rises in the BE simulation. In common with LBS and NIESR, both HMT and BE also allow for an effect of capacity utilisation on wholesale prices with similar short-term effects on the price level but, in contrast, this pair of models define the change in capacity utilisation to be temporary for a given shock to output. The LBS and NIESR models have no prior constraints on the duration of changes in capacity utilisation, but these emerge from the dynamics of output and capital stock adjustment. Very protracted adjustment of the capital stock in the LBS model implies a steady rise in capacity utilisation following a demand shock, and a slow return to base levels of utilisation for the NIESR model. In the full model simulation where output continues to change, excess capacity is created in the last two years of the NIESR simulation as GDP effects are reversed. Clearly there is some dispute between the models over the response of utilisation to output shocks, and the sustained change produced by the LBS model appears to be the least plausible account.

The Public Sector Borrowing Requirement

In general the net PSBR cost of any change in fiscal policy is likely to be less than the gross cost, to the extent that the policy stimulates output and employment and generates flowback from higher tax revenues and lower unemployment-related benefits. A further effect arises from the influence of inflation on tax revenues relative to that on expenditures. In each of the tables of fiscal policy simulation results, in addition to the net PSBR cost of the policy measure, we also show the gross cost (in parentheses) both in billion [pounds] 1989 prices, the change in the PSBR being deflated by the GDP deflator so that the figures are comparable over different years. This allows the reader to standardise the different fiscal policy simulations according to either their net or their gross cost and so assess their relative cost effectiveness. In the government expenditure simulation, where the gross cost of the policy change is 2 billion [pounds] (1989 prices) the net PSBR cost is some 40 per cent of this in the first year of the NIESR, HMT and BE simulations. In the HMT simulation the net cost falls slightly over the simulation, to about 50 per cent of the gross cost, whereas in the NIESR and BE simulations this ratio falls steadily to about 20 per cent. In the LBS model the net cost becomes negligible by the third year: this implies that the policy is self-financing. This greater degree of flowback on the LBS model is largely explained by the greater stimulus to output relative to the other model simulations.

3. Other policy simulations The direct effect of tax cuts on real consumption wages The following three simulations examine the effects of cuts in income tax, VAT and employers' NICs. Although for the various models many of the features of these simulations resemble those of the government expenditure simulation, an important difference is that all of the tax changes can potentially exert a direct influence on wage determination. The extent of this influence depends critically on the specification of the model wage equations and we therefore begin by examining the role of taxes in the long-run solution to these equations in each of the models. The long-run solutions to the model wage equations can be represented as: w = p + [Alpha.sub.1][t.sub.d] - [Alpha.sub.2][t.sub.c] - [Alpha.sub.3][t.sub.e] + other terms [Mathematical Expression Omitted] where w is the log of nominal average earnings, p is the log of consumer or retail prices and [t.sub.d], [t.sub.c] and [t.sub.e] are the direct, indirect and employers' tax rates (expressed in percentage form) respectively. The values of the coefficients of [Alpha.sub.1], [Alpha.sub.2] and [Alpha.sub.3] for each of the quarterly models are shown in table 8 below. To the extent that a cut in taxation increases real consumption wages, (w--p--[t.sub.d]), there will be an increase in real disposable income and hence consumers' expenditure and aggregate demand through the usual multiplier process. Conversely, the smaller the rise in real consumption wages the more the tax cut will be reflected in a fall in real wage costs (earnings plus employers' contributions deflated by some measure of output prices). The transmission mechanism by which any resulting change in real wage costs affects the economy depends on a variety of factors, which differ widely across the models. One route is via the demand for labour, where any impact of a fall in real wage costs depends on the price elasticity of demand for labour. In the models the elasticity of demand for manufacturing employment with respect to real wage costs is -0.3 for NIESR, -0.4 for HMT, -0.5 for LBS, and -0.6 for BE, implying that a reduction in unit labour costs of 1 per cent leads to approximately 15-25 thousand more manufacturing jobs. There is greater dispersion in the elasticities of non-manufacturing employment, ranging from zero for NIESR, -0.05 for HMT, -0.2 for BE to -0.5 for LBS, so that a reduction in unit labour costs of 1 per cent has a direct effect on non-manufacturing employment somewhere between 0 and 60 thousand. A decline in unit labour costs can also directly affect manufacturing investment in the NIESR and HMT models by changing relative factor costs or by improving company profitability. Through downward pressure on inflation, a cut in unit labour costs can also stimulate demand indirectly through a range of channels. One of these is the inflation effect on consumption, discussed above; another is an improvement in competitiveness in domestic prices, which stimulates net exports.

Income tax cuts

In discussing the effects of an income tax cut on the previous vintage of models we highlighted the importance of `retention ratio' effects. 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. Both HMT and BE incorporated a retention ratio effect in the previous vintage of models (i.e. [Alpha.sub.1][is greater than] 0 in equation 1), whereas two, NIESR and LBS, did not (i.e. [Alpha.sub.1] = 0). Turner et al. (1988) found that for this earlier vintage of models the statistical evidence for the inclusion of the retention ratio effect was weak. In the current set of models there is now a retention ratio effect incorporated in the wage equations of HMT, BE (having re-estimated the system of wage equations in response to points raised by Turner et al., 1988) and NIESR (who argue that the evidence for its inclusion is now `more clearcut'), so that the LBS model is alone in not including a retention ratio effect. The HMT and NIESR earnings equations imply that a 1 per cent increase in the retention ratio will eventually lead to a reduction in the level of earnings by about 0.6 per cent and 0.4 per cent respectively (table 8). In the BE model the effect differs between sectors with wages falling by about 0.6 per cent in the manufacturing and public sectors but only about 0.3 per cent in the private non-manufacturing sector.

In the full model simulations the presence of the retention ratio effect is most obvious in the case of HMT and BE, leading directly to a reduction in unit labour costs so that there is no rise in consumer prices over the first three years of the HMT simulation, and a steady decline in consumer prices in the BE simulation, despite increases in GDP and employment. The continuing fall in consumer prices in the BE simulation occurs because, in the absence of a marked exchange-rate response, counteracting demand-pull effects on prices are comparatively weak. Exchange-rate effects are also weak for HMT but demand-pull forces, particularly acting through the effect of falling unemployment on wages, eventually lead to some increase in prices. However, the inflationary cost of reducing unemployment by cutting income tax for this model is still substantially lower than that of increasing government expenditure.

The importance of the retention ratio is not so immediately apparent in the NIESR simulation which has many of the same characteristics as the NIESR government expenditure simulation: a step fall in the nominal exchange-rate of about 3 1/2 per cent followed by rapidly increasing prices with a peak in the GDP response after three years. However the downward pressure on unit labour costs as a result of the retention ratio effect means that the speed and extent to which prices increase is significantly less than for the government expenditure simulation, despite the fact that the step fall in the exchange rate is of a similar magnitude. This in turn means that the extent to which GDP is crowded out is significantly less than for the government expenditure simulation.

Despite the fact that the LBS model does not incorporate a retention ratio effect, the GDP effects are actually greatest in the LBS simulation, a result which is explained by the large initial fall in the exchange rate and sluggish price adjustment (similar to the government expenditure simulation). However, perhaps a more meaningful comparison is that the LBS is the only model for which the government expenditure simulation compares favourably with income tax cuts in terms of the inflationary cost of increasing GDP.

Cut in employers' National Insurance contributions

The largest GDP effect occurs in the NIESR simulation, with GDP around 1 per cent higher after the second year. These favourable results arise because inflationary pressures are more contained as a direct effect of the lower real wage costs, since (as for LBS) the tax change is borne entirely by employers, with no direct effects on real consumption wages (i.e. [Alpha.sub.3] = 0). Indeed, of all the NIESR simulations, the cut in employers' NICs produces the most sustained rise in GDP with the smallest inflationary cost. Higher investment results both from an accelerator response to the more sustained improvement in domestic output, and in the case of manufacturing investment, to an improvement in company profitability. The increase in manufacturing investment raises manufacturing capacity by 0.9 per cent after five years (higher than for any other NIESR simulation) which also helps to restrain inflation. This means that there is some positive contribution from consumers' expenditure to the multiplier process. Increased manufacturing capacity also directly reduces manufacturing imports and improves the trade balance.

A strong contribution from net trade is also a feature of the LBS simulation, as the competitive gains from the characteristic fall in the exchange rate are further enhanced by the fall in unit labour costs, with exports accounting for most of the increase in total final expenditure and imports actually falling despite the increase in GDP of nearly 1/2 per cent. The absence of any strong growth in investment, which just increases in line with output, is due to the absence of any direct link between profitability and investment in the LBS model. As for NIESR, consumer prices do rise in absolute terms mainly because of the fall in the exchange rate and higher import prices.

The increase in GDP in the HMT simulation is of a similar magnitude to LBS, rising gradually to nearly 1/2 per cent after five years, whereas in the BE simulation the GDP effects peak after three years at about 0.3 per cent, with little effect by the fifth year. The composition of the change in expenditure sets the BE and HMT simulations apart from NIESR and LBS, as it is mostly concentrated in consumers' expenditure. The initial increase in consumers' expenditure is a result of higher real disposable income, in turn caused by short-run upward pressure on real consumption wages (as [Alpha.sub.3][is greater than] 0 in the short run for HMT and BE) and a lag in the adjustment of nominal average earnings to lower prices. Nevertheless the cut in employers' NICs does result in a fall in real wage costs with downward pressure on domestic prices. In both the HMT and BE models this is translated into a corresponding appreciation of the nominal exchange rate so that there is a continued fall in domestic prices, rather than an improvement in competitiveness, and hence net exports, as in the LBS and NIESR simulations. This fall in domestic prices further stimulates consumers' expenditure. Investment responds passively to the increase in output in the BE model since there is no link between increased profitability and investment, and in the HMT model the change in relative factor prices outweighs the effect of any improvement in profitability on manufacturing investment. The downturn in the GDP effect in the BE simulation occurs because in the long run there is no direct effect of employers' NICs on real consumption wages ([Alpha.sub.3]=0) and without any strong transmission mechanism to stimulate either investment or net exports there is instead an accelerating fall in domestic prices. However, employment rises in response to lower real wage costs, producing a fall in unemployment which is comparatively large to that observed in other BE simulations. Cuts in employers' NICs (together with cuts in VAT) stand out as the most effective means of reducing unemployment in terms of their inflationary cost on both the BE and HMT models.

The effect of a cut in VAT

The direct effect of a cut in VAT is to reduce the price level in all the models as almost all the change in VAT is passed on to the consumer in the NIESR, HMT and BE models. There is a weaker direct effect in the LBS model because the VAT rate enters the equation for consumer prices in logarithmic form, which counter-intuitively implies that the higher the initial level of the VAT rate, the smaller the effect on the price level of any given percentage point change in the VAT rate. This feature explains why the LBS model is the one in which a cut in VAT does not imply a lower inflationary cost of increased GDP in comparison to an increase in government expenditure.

In the HMT and BE simulations there is no direct effect of a cut in the VAT rate on real consumption wages ([Alpha.sub.2] = 0), although in the HMT model there is upward pressure on real wages as a result of increased company profitability, and in both models there is some short-run increase in real earnings as nominal earnings lag behind the fall in consumer prices. Nevertheless average earnings do follow the decline in consumer prices so that the cut in VAT is reflected in a decline in unit labour costs. Thus the effects of a cut in VAT on the HMT and BE models broadly resembles those of a cut in employers NICs, the main difference arising because the former have a much more immediate impact on the price level.

In the NIESR simulation a cut in VAT leads directly to upward pressure on real consumption wages ([Alpha.sub.2] = 0.6), although since [Alpha.sub.2] [is greater than] 1 this still allows for some decline in real wages. Thus the effects of a cut in VAT are not the same as a cut in employers' NICs where there is no direct effect on real consumption earnings. However, as noted by Wren-Lewis (1988), the restrictions imposed on the NIESR earnings equation, namely (1-[Alpha.sub.1]) = [Alpha.sub.2], imply that over the long run the effects of a cut in VAT will be broadly similar to those of a cut in income tax. Nevertheless over the short run the cut in VAT does have significant advantages over the income tax cuts which again stem from its more immediate impact on the price level, and which in turn improves the GDP response through channels discussed above.

Cut in short-term nominal interest rates

In this exercise we consider the effects of reducing short-term interest rates by 1 percentage point. In the BE, LBS and NIESR models short-term interest rates are usually determined by a policy reaction function which in the present exercise is over-written. There is typically some form of term structure equation relating long rates to short rates, and these are assumed to continue to operate. In the LPL model interest rates are jointly determined with the exchange rate to maintain uncovered interest parity and hence this experiment is not performed on 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 simulation 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 (or `single-equation') effects are calculated from the individual equations which determine these variables, assuming that all other variables (except interest rates) remain unchanged. The full simulation results shown in table 5 are then the sum of these direct effects plus indirect effects which occur from feedback through all the model equations.

In table 9 we show the direct effects of a 1 percentage point cut in interest rates on consumption (excluding any indirect effects via wealth terms), investment and stockbuilding after five years, giving both the absolute change (at [pound] m 1985 prices) and the proportionate change for each expenditure aggregate. The greatest direct expenditure contribution to GDP for LBS, HMT and BE is from consumers' expenditure. The response is weaker in the NIESR model and is dominated by that from total fixed investment (primarily investment in distribution and business services, and private housing investment). There are no direct interest-rate effects on non-manufacturing investment in the HMT and BE models, but in the former the impact on manufacturing investment is at least twice as large as any other model. There is a substantial increase in stockbuilding in the short run across the models (of similar order to the rise in fixed investment) but this is less influential after five years, given that for most of the models it is the level of stocks that depends on interest rates (usually through a cost of stockbuilding term). However, of note is the stockbuilding contribution in the HMT model where the lags on interest rates are relatively long, and distributors' stocks in the LBS model, where there is a particularly large interest-rate effect.

In total, the direct expenditure effects for LBS dominate those for the other models, with HMT containing the next largest effects and with a small effect from NIESR (in sharp contrast to the previous version of this model). Differences in the response of exchange rates also add to the divergence between models. In the HMT model the direct effect of every 1 percentage point cut in interest rates is to lower the real exchange rate immediately by about 1.3 per cent, whereas for the BE model a change in interest rates has a much weaker effect with the real exchange rate falling by only 0.3 per cent after a lag. In the NIESR model a fall in interest rates of 1 percentage point which is expected to last for five 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.

In the full model simulations the increase in GDP in the first year is about 0.4 per cent for LBS, NIESR and HMT, with the direct effects on the expenditure components described above being further supplemented by the effect of the exchange rate. Because the direct effects on the expenditure components, particularly consumption, are comparatively weak in the NIESR model, the fall in the exchange rate is important in stimulating exports and stockbuilding. But given the deterioration in competitiveness and the nature of the interest-rate cut for the NIESR model, there is a reversal of the improvement in GDP in the final year of the simulation. In the LBS simulation, where the shock is permanent, and hence the direct effects of higher interest rates persist, the effects of the deterioration in competitiveness are less marked. In the HMT simulation the fall in the exchange rate, and hence the stimulus to net trade is smaller; nevertheless there is also a strong income effect on consumption largely due to a fall in personal sector interest payments, particularly mortgage repayments, relative to interest-rate receipts. The magnitude of this effect in the first year of the HMT simulation is almost as large as the direct substitution effects, and although it diminishes in later years as lower interest rates also encourage increased personal sector borrowing it helps to maintain this model's strong GDP response. In the BE simulation where the direct effects on consumption and investment are relatively weak and the fall in the nominal exchange rate is also small, the resulting impact on GDP in the short run is much weaker than in the other models. However in the second year, the increase in GDP more than doubles as a consequence of multiplier and accelerator effects, so that the increase in GDP is of a similar order of magnitude to that observed in the other models.

By the fifth year of the simulation the greatest increase in prices, over 3 per cent, is in the HMT and NIESR models, where the fall in the exchange rate is greatest. By contrast the increase in consumer prices is only 1/4 per cent by the fifth year of the BE simulation, reflecting not only the relative insensitivity of the exchange rate to changes in interest rates, but also the absence of any strong demand-pull pressures in this model. The price effects reported in the results tables relate to the consumer price index (CPI) although for most of the simulations the retail price index (RPI) follows a virtually identical path. However, in the case of a change in interest rates these two price indices diverge, as only the latter includes a component for mortgage costs. Across all of the models the percentage change in the RPI is at least 0.3-0.4 per cent lower than the percentage change in the CPI over the entire simulation.

A reduction in unemployment benefits

In this simulation the rate of unemployment benefit and the rate of supplementary benefit paid to the unemployed are cut by 10 per cent in real terms, which according to figures in the latest Public Expenditure White Paper should give a gross saving to the PSBR of about 0.5-0.6 billion [pounds] (1989 prices). In the case of LPL it is assumed that the revenue effects are re-distributed as lump-sum transfers, as the model is dynamically unstable if the savings are assumed to reduce the PSBR.

A cut in unemployment benefit has both demand-side and supply-side effects. On the demand-side, a cut in benefits reduces personal disposable income and hence consumption and aggregate demand via the multiplier process. Supply-side effects arise from of the possible influence of unemployment benefits on the determination of wages and/or on the decision to seek work or register as unemployed.

For the quarterly models the simulation results show that the demand-side effects are responsible for an overall reduction in GDP. In the case of NIESR and BE there are no supply-side effects from a cut in benefits, NIESR having argued in the past that they have been unable to find any empirical support for such effect). The fall in GDP then results in a fall in employment and a rise in unemployment, as might be expected following any reduction in aggregate demand. However in the LBS, HMT and LPL models supply-side effects are present, although the transmission mechanisms through which they operate differ across the models.

In the case of the LBS model, a cut in unemployment benefit of 10 per cent reduces real non-manufacturing wages by about 1 per cent, which in turn has some effect on increasing the demand for labour in the non-manufacturing sector. The resulting simulation results show a very slight overall increase in employment of about 5-15 thousand, and a fall in unemployment of a similar magnitude, as these supply-side effects marginally outweigh the lower labour demand resulting from the reduction in GDP. In the HMT simulation a cut in benefits discourages the unemployed from registering and so reduces the unemployment count by about 90 thousand in the first year. But as the demand-side effects of lower GDP build up, about half of the initial fall in (registered) unemployment is cancelled out and the net result in the HMT simulation is thus a decline in both employment and unemployment of about 45-50 thousand. The results in the LPL simulation are very different from the other models, with a substantial increase in GDP and employment, and a large fall in unemployment and inflation. In the LPL model unemployment benefits enter into the determination of real wages with a long-run elasticity of about 0.5, implying that a 10 per cent fall in unemployment benefits will, ceteris paribus, eventually result in a 5 per cent fall in real wages, although the empirical basis for this equation is disputed (see Nickell, 1984, and Minford, 1985). In the full model simulation this leads to a fall in the equilibrium level of unemployment of about 20 per cent and a rise in the equilibrium level of GDP by 2 1/4 per cent. These effects put upward pressure on real wages so that the resulting fall in real wages is only about 2 per cent. The absolute fall in unemployment of about 200 thousand is significantly less than for similar simulations on the LPL model in the past (see, for example, Wallis et al., 1987). These results merely represent the same percentage change in unemployment and reflect the lower level of unemployment in the base forecast.

4. Conclusions The simulations presented in this article are chosen to illustrate model responses to key policy instruments. The presentation has concentrated on the response of the main macroeconomic aggregates, and the tabulated results represent ready-reckoners that can be used in further policy analysis. Other summary statistics can be derived from the simulation results at the reader's discretion; for example, the inflation cost of reducing unemployment or the comparative PSBR or balance of payments costs per job.

In our previous model review, we highlighted particular mechanisms which contributed to the differences in simulation results between the models. Further investigation of the labour market specifications (Turner et al., 1988) showed that several such differences were capable of resolution. Some of this research is incorporated in the current versions of the models, although some of the critical differences remain (for example, the role of wealth in the LPL durables equation and the treatment of short-term unemployment in the BE model). Overall, one might expect to observe a greater degree of conformity in simulation results over time. However the current models incorporate some radical changes in approach and simulation differences between them have widened.

Of particular note are the changes in the structure of the NIESR model leading to marked changes in simulation responses. One important change is the adoption of a new exchange-rate equation which contributes to the greater inflationary pressure observed in simulations. Another is the introduction of a vintage production technology which provides explicit estimates of capacity utilisation; this measure of demand pressure in the goods market also explaining part of the greater inflationary response of this model. These changes are reinforced by the speed of adjustment of prices. The new NIESR model also contains an important role for wealth in the determination of consumption, and hence output, but the strong adverse influence of inflation on wealth makes this model somewhat of an outlier in this respect. Many of the simulation results of this model, notably that of government expenditure, indicate its movement away from the simple Keynesian position.

The LBS model has also developed in the spirit of incorporating explicit supply considerations, in particular by the inter-related nature of output and price determination and the demand for factors of production. However, model simulations reveal surprisingly little difference from the previous version of this model. Several factors appear to contribute to this. Of particular note is the lack of homogeneity in the price system. Very long lags on capital stock adjustment not only generate sustained medium-term changes in capacity utilisation, but also produce much more marked employment increases in relation to a given change in output than for the other models. As a consequence the output effects of a demand expansion remain substantial.

There has been less revision to the HMT and BE models and hence comparatively little change in their simulation properties. The BE model continues to produce much less inflation than do the other quarterly models. Not only are pressure of demand effects in both labour and goods markets weak, but there is also little contribution from exchange-rate movements. In contrast, labour market pressures in the HMT model are reinforced by strong feedback effects.

In the HMT and BE models of consumption potentially powerful inflation effects on the income variable are offset by the influence of inflation on real interest rates. Consumer behaviour is an area of much current research, although the results presented here suggest that model differences may not be resolved solely by examination of the behavioural equations determining consumption, and that it is also necessary to analyse the response of wealth and personal sector net holdings of liquid assets to changes in activity and inflation.

The simulation results demonstrate that the effects of changes in different taxes depend critically on the specification of the wage equations. Since across all of the quarterly models a cut in employers' NICs is reflected in a fall in real wage costs rather than a rise in real consumption wages, this simulation produces falls in unemployment at a comparatively low inflationary cost. The absence of any direct effect of VAT on real consumption wages in the BE and HMT models means that over the long run the effects of a cut in VAT and employers' NICs are broadly similar. The presence of a retention ratio effect continues to be an important source of difference in the income tax simulations. Although Turner et al. (1989) found that the statistical support for such an effect was not strong in the previous vintage of models, the inclusion of a retention ratio effect in the latest version of the NIESR model (where previously there was none) suggests that this issue has yet to be resolved.

Exchange-rate differences continue to play an important role in explaining some of the disparity in model simulation responses. The models all share a reasonably common background to the determination of exchange rates but practical differences are important in influencing different patterns of behaviour. Fisher et al. (1989) show that it is possible to derive a 'consensus' exchange-rate equation for three of the quarterly models which is at least as good as any of the existing equations in a statistical sense; substituting this new formulation into the models considerably narrows the differences in the response of output ot a demand shokc. In the case of LBS the exchange rate is a product of the complete financial sector and hence it is not possible to analyse the precise mechanism leading to a change in this variable. We note that Courakis (1988) has been critical of many features of the LBS financial sector.

Although continued differences between the models might imply a lack of policy reliability, it is clear that at least some of the differences between the current models reflect attempts to incorporate features hitherto neglected. There are three main factors which appear to have motivated these developments. First, the relatively large swings of output over the last decade have not only highlighted the desire to explain both the contribution to, and the response of, the supply side of the economy, but have also provided a more informative environment in which these influences can be empirically assessed. Second, the increased emphasis on supply-side policy has induced a desire to analyse and evaluate the current impact and likely future success of these policies. Finally, the developments would seem to reflect a desire for greater theoretical consistency in the models, both in respect of a better articulated macroeconomic framework whereby both demand and supply factors are incorporated, and also in terms of internal consistency, for example the joint determination of output, prices and the demand for factors of production. Whether these innovations prove to be empirically robust has yet to be fully judged. [Tabular Data 1 to 9 Omitted]

P.G. FISHER, S.K. TANNA, D.S. TURNER, K.F. WALLIS and J.D. WHITLEY are members of the research team at the ESRC Macroeconomic Modelling Bureau at the University of Warwick. A PC-program which uses many of the results described in this paper in ready-reckoner form is available from the Bureau to academic institutions at a small charge. Editorial responsibility is taken by the authors, not by the Editorial Board of the Review.
COPYRIGHT 1989 National Institute of Economic and Social Research
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1989 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Fisher, P.G.; Tanna, S.K.; Turner, D.S.; Wallis, K.F.; Whitley, J.D.
Publication:National Institute Economic Review
Date:Aug 1, 1989
Previous Article:Routes to economic integration: 1992 in the European Community.
Next Article:The economic situation.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters