Comparing global economic models

Comparing global economic models

Economic Modelling 15 Ž1998. 1]48 Comparing global economic models Peter R. Mitchell, Joanne E. Sault, Peter N. Smith, Kenneth F. WallisU ESRC Macroe...

226KB Sizes 1 Downloads 26 Views

Economic Modelling 15 Ž1998. 1]48

Comparing global economic models Peter R. Mitchell, Joanne E. Sault, Peter N. Smith, Kenneth F. WallisU ESRC Macroeconomic Modelling Bureau, Uni¨ersity of Warwick, Co¨entry CV4 7AL, UK

Received 9 February 1996; accepted 6 January 1997

Abstract This article presents a first account of comparative research on multi-country models conducted by an independent third party with full hands-on access to the models. The models are MSG2, MULTIMOD and NIGEM. Their structural characteristics are first compared and contrasted; their theoretical foundations are seen to have some broad similarities. Overall model properties are then elucidated by three simulation experiments conducted in a standardized policy framework. The source and nature of observed crossmodel and cross-country differences in the results are described, a fundamental source being the model-builders’ different approaches to quantification. Q 1998 Elsevier Science B.V. JEL classification: C5 Keywords: Multi-country models; Model comparisons; Simulation; Rational expectations

1. Introduction Macroeconometric models, national or international, play a key role in the policy-making process. They provide a consistent and comprehensive account of the relevant economic interactions and interdependencies, quantified with referU

Corresponding author.

0264-9993r98r$19.00 Q 1998 Elsevier Science B.V. All rights reserved P I I S 0 2 6 4 - 9 9 9 3 Ž 9 7 .0 0 0 1 5 - 1


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

ence to historical data. National economy-wide models typically treat economic conditions in the rest of the world as exogenous, whereas macroeconomic interactions among nations are a major focus of global models, which have been constructed and used in a wide range of international official and academic institutions. Different models are built for different purposes, however, and not only models but also the problems they address evolve over time, as social and political changes alter the focus of economic attention. Moreover models are built in accordance with an underlying view of the way the world works. Thus different models may provide different estimates of the response of the economy, national or international, to policy interventions and innovations. It is then important to identify the sources of such disagreement and to reduce it where possible. Several multi-country model comparisons have been undertaken in recent years. Particularly noteworthy are the projects sponsored by the Brookings Institution ŽBryant et al., 1988, 1993. and the European Commission ŽBarrell and Whitley, 1992.; in each case the cited volume reports the proceedings of a conference with which the project culminated. These followed the style of many model comparison conferences that have been held over the years: an organizing group determines the broad outlines of exercises to be undertaken on a number of models; model proprietors attempt to carry out these exercises on their own models; the results are presented and discussed at a conference; the proceedings are published. Although the Brookings and EC-sponsored projects stand out by virtue of the detailed attention given to experimental design ex ante and the depth of comparative analysis by third parties ex post the execution of the experiments remained in the hands of the model proprietors and complete standardization was not achieved. Thus, when important differences are observed in the models’ predictions of the global economic response to policy interventions of various kinds, as they invariably are, it may be that these are due at least in part to differences in side conditions or other adjustments that individual model proprietors have made, but the extent to which this is true remains unknown. More generally those analysing the results have not had access to the models to undertake further investigations, hence little explanation of perceived differences has been provided, except in terms of the overall perspective taken by the different models on rather broad questions, such as the treatment of expectations. A different and, as yet, unique style of model comparison project is represented by the comparative research programme of the ESRC Macroeconomic Modelling Bureau. As an independent third party with whom complete models and associated databases are deposited, the Bureau is able to undertake direct comparisons across competing models at all stages of the project } design, execution, analysis and testing: Wallis Ž1993. contrasts this with the more common style of model comparison conference. This article presents a first account of comparative research on multi-country models carried out in the same manner, the Bureau’s research having hitherto been primarily focused on models of the UK economy. Three of the models that participated in the Brookings exercises and have been made available to us by their proprietors form the basis for this pilot project:

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48 v




MSG2 ŽMcKibbin]Sachs Global. model wsee McKibbin and Sachs Ž1991.x, revised and maintained by Warwick McKibbin at the Australian National University and the Brookings Institution; MULTIMOD, developed in the Research Department of the International Monetary Fund Žsee Masson et al., 1988, 1990.; and NIGEM, developed at the National Institute of Economic and Social Research Žcurrently under the supervision of R.J. Barrell. and jointly maintained with the London Business School.

All three are implemented at the University of Warwick, with MSG2 and NIGEM running in their own software and MULTIMOD having been adapted to run in the Bureau’s model-handling system. The content of the remainder of this article is as follows. In Sec. 2 we describe the structural characteristics of the models, which rest not only on the theoretical foundations assumed by each modeller but also on their different approaches to quantification. Three key areas compared and contrasted are wage determination, consumption and investment decisions and, both in these areas and more generally, the nature and extent of forward-looking behaviour. In Sec. 3 we describe the overall properties of the models as revealed in three simulation experiments carried out under standardized conditions. This means standardizing the fiscal closure rule and the monetary policy assumption, and the experiments comprise a fiscal shock, a monetary shock and an external shock. Various cross-model and cross-country differences that emerge in the results are explained in terms of structural features elucidated in the preceding section. Our hands-on access to the models allows variant simulations to be undertaken which further illuminate some of these issues. Sec. 4 contains concluding comments.

2. Three global models 2.1. General characteristics Although the models describe the same global economy they differ in respect of which countries are modelled individually and how the remaining countries are grouped together. All three global models contain separate models for each of the Group of Seven countries: the US, Japan, Germany, UK, France, Italy, Canada. MULTIMOD then divides the rest of the world into three groups } smaller industrial countries; high-income oil exporters; and developing countries } and is the most aggregated of the three models. The version of the MSG2 model we are using ŽEuropean version 36. divides the rest of the world into five groups } the remainder of the EMS countries Žthe Benelux countries, Denmark and Ireland., the remainder of the OECD member countries, oil-exporting countries, non-oil developing countries and Eastern Europe including the former Soviet Union. The NIGEM model is the most disaggregated of the three, first adding Spain, the Netherlands and Belgium to the list of countries having their own submodel. Next


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Denmark, Ireland, Greece and Portugal are aggregated into a ‘remainder of EC’ model, Sweden, Finland, Norway, Iceland, Switzerland and Austria form the ‘rest of Europe’, and Australia and New Zealand are aggregated to complete the coverage of the OECD countries. Finally, while Eastern Europe and the bloc of oil producers appear as in MSG2, in the NIGEM model some LDCs are captured through a regional approach, thus separate descriptions of the economies of Africa, Latin America, the Far East and a remainder complete the system. In all cases the most important groups of countries, the major industrialized nations, are fully modelled with an internal macroeconomic structure that has a common theoretical foundation across countries, whereas in the other groups of countries only the foreign trade and external financial aspects are represented, in a more reduced form than structural manner. The theoretical foundation also has common elements across the three models, all of which follow the prevailing paradigm in which a broadly neoclassical view of macroeconomic equilibrium coexists with a new Keynesian view of short-to-medium term adjustment. This represents an extension to the simple Mundell]Fleming model, which is the basic underlying model of interdependence, by adding dynamics and modelling the processes generating aggregate supply; accounting for the behaviour of asset and debt stocks as well as financial flows is also undertaken. Finally all three models incorporate forward-looking behaviour in financial markets and, in some cases, in decision-making by firms and households, and they all treat future expectations as ‘rational’ or ‘model-consistent’ expectations. Some important differences across models andror countries that nevertheless appear are described below. Different approaches to quantification represent an important potential source of such differences. The MSG2 model is a dynamic general equilibrium model and is calibrated in accordance with the practice of ‘computable general equilibrium’ ŽCGE. modellers. Some parameters are assigned values, common across countries, taken from published econometric studies; others correspond to observed expenditure shares or input]output coefficients in a benchmark year, assumed to represent an equilibrium of the model. Only in the specification of the model’s wage]price dynamics are there important behavioural differences between regions, discussed below. In MULTIMOD and NIGEM, on the other hand, parameters are estimated from historical data by the conventional techniques of time-series econometrics. In MULTIMOD, pooled regression techniques are employed to estimate common parameter values across countries and there are few countryspecific behavioural equations. In NIGEM a common theoretical structure is estimated separately for each country, and different dynamic adjustment patterns and parameter values are allowed to emerge from the data. In this case these are quarterly, in contrast to MSG2 and MULTIMOD, whose time unit is a year. 2.2. Wage formation The question of whether there are significant differences in the structure of the aggregate labour market between countries has been examined by a number of authors. A useful benchmark is provided by Layard et al. Ž1991., who report

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


estimates of a common structure of wage, price, employment and unemployment determination for a wide range of countries and compare their results with those of others. In particular they find significant differences in the dynamics of wage determination across countries. Their results indicate that Japan shows very little real or nominal wage rigidity, whereas both are substantial in Germany, France and the UK, with the US occupying an intermediate position. Subsequently Turner et al. Ž1993. report similar findings for the US, Germany and Japan. The NIGEM model adopts the same empirical specification strategy as these authors, as noted above. The general form of wage equation that is estimated separately for each country is as follows e Dwt s a 1 Ž t ptqi y pt . q a 2 D pty1 q a 3 Ž a 4 z ty1 y wty1 . q a 5Ut


e where w is the nominal wage, p the price of output, t ptqi the expected price level in period t q i based on information in period t, z a set of variables determining the long-run level of wages and U the level of unemployment or similar cyclical variable; D is the first difference operator. All variables except U are assumed to be in logarithms. The theoretical basis of this model can be traced to Fischer Ž1977. and Taylor Ž1979., where the dynamics of the model and the resulting wage stickiness are motivated by reference to overlapping wage contracts. Without giving the full estimation results, the key coefficient values are as follows. In the cases of Japan and Germany a 1 s 0, expected future prices playing no significant role. In the US a 1 s 0.496 and i s 4, implying that 1-year-ahead expectations are significant. In Japan, Germany and the US a 2 s 1.54, 0 and 0, respectively. Thus, while future expectations of prices are absent for both countries, price inflation is more important in Japan than in Germany. Finally, a 5 s y0.0196,y 0.00915 and y 0.00121 for Japan, Germany and the US. Taken together, these coefficients imply that wage inflation in Germany is more sluggish than in the US and in Japan. The same finding is represented in the MSG2 model by using three versions of a wage equation which can again be nested in Eq. Ž1.. The first, for all countries and country groupings except Germany, the REMS and Japan, is e Dwtq1 s a 1Ž t ptqi y pt . q Ž 1 y a 1 . D pt q a 5 Ž l t y l t .


where l t is employment and l t its full employment level. Wages are set one period ahead, with the degree of forward-looking behaviour controlled by the coefficient a 1 , whose typical value is 0.4. The coefficient a 3 in Eq. Ž1. is set equal to zero which implies that the long-run level of wages is not determined through this equation but rather through the full model solution. The real wage in MSG2 is determined by the marginal product of labour in the long run, that is, directly through the output supply process. For Germany and the REMS, l t is replaced by ) l t) , the short-run natural level of employment, where D l t) s 0.2Ž l ty1 y l ty1 . q U 0.7Ž l ty1 y l t .. In the long run l converges to full employment l, an exogenous variable. This specification produces a large degree of persistence in unemploy-


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

ment in Germany and the REMS and, in particular, reduces the flexibility of wages when subject to a nominal shock. In Japan, by contrast, wages are assumed to be sticky for one period only, that is they are set one period ahead at their expected equilibrium values. This is achieved through the equation wtq1 s ptq1 q g Ž l tq1 y l t) . .


This specification implies that Japanese wages are at their equilibrium value two periods after a shock takes place, which represents a strong degree of flexibility when compared to the specification for other countries. Output prices in MSG2 are modelled as equal to the marginal cost of output, which is equal to average cost given the Cobb]Douglas nature of production. Given that there are no short-run dynamics in the calculation of marginal cost, all of the dynamics come from the process of wage setting through the above equations. In MULTIMOD wage setting is subsumed into the price setting process and, in the absence of a fully articulated model of output supply, price setting is through an expectations-augmented Phillips curve. This ‘semi-reduced-form’ equation is again intended to capture wage setting in a framework of overlapping contracts. The estimated equation employed in the models of the G7 economies is D pt s 0.294D ptq1 q 0.400D pty1 q 0.306D pt) q 0.285CUt


where p is the non-oil GNP deflator, pU is the price of domestic absorption and CU is capacity utilization, a measure of the level of excess demand Žprice variables in logarithms.. The International Monetary Fund Ž1991. details the econometric results, including tests for region-specific values for the parameters. The evidence presented fails to reject the restriction of common parameters, although the unrestricted estimates ‘suggest that the response of prices to current market conditions... may be greater in Japan’. The evidence also rejects restricting the equation to be entirely backward-looking. The common specification across countries implies that MULTIMOD is an outlier in this respect. 2.3. Consumption and in¨estment decisions The underlying model of aggregate consumption in MSG2 and MULTIMOD is the life-cycle model, based on an assumption of intertemporal optimization, in which consumption is proportional to total wealth, comprising human and financial wealth. Human wealth is defined as the present discounted value of the flow of future after-tax labour income, a definition which is made operational by expressing current human wealth as a function of its next-period discounted value and current income. However, only a proportion of households are assumed to implement full intertemporal optimization; the remainder are unable to carry out these plans, perhaps because of liquidity constraints, and consume out of current income. The proportion of ‘life-cycle’ households is set at 0.3 in all countries that are fully modelled in MSG2. In MULTIMOD the estimated consumption function based on these considerations is in error-correction form, with consumption homogeneous of

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


degree one in wealth in the long run, the ratio of consumption to wealth depending on the Žreal, long-term. interest rate. The pooled estimate across the industrial countries, neglecting the intercept term, an intercept shift and a demographic variable, is D c t s 0.150 Ž Wty1 y c ty1 . q 0.330D yt y 0.154RLR t


where c, y and W are Žlog real. consumption, disposable income and total wealth, respectively, and RLR is the real long interest rate. Thus consumption falls in the long run by approximately 1% for a 1 percentage point rise in the interest rate. NIGEM has a more eclectic approach, the estimation strategy noted above allowing some rather different responses to emerge across countries. Again using an error correction form, with up to five lags in quarterly data, consumption is a function of personal disposable income, net financial wealth, nominal interest rates and inflation. A long-run unit elasticity on income and wealth is accepted and imposed in all cases, although differences occur in respect of the remaining variables. Of the G7 countries, for example, two have a direct long-run interest rate effect on consumption, two have an inflation effect and three have neither. With respect to their approach to the modelling of investment behaviour, a similar division of the three models occurs. The underlying theoretical framework in MSG2 and MULTIMOD is again that of intertemporal optimization, in which firms invest up to the marginal valuation given by Tobin’s q. Investment then responds to a change in the real Žshort-term. interest rate via its effect on the market value of the capital stock. In the MSG2 model, however, a proportion of firms is unable to borrow and lend as the theoretical model assumes, and invests out of current profits. The NIGEM model disaggregates more, distinguishing housing investment, business investment and stockbuilding, which is probably more useful in forecasting, and the underlying models are then more traditional. Housing investment is treated in a similar manner to consumption, while business investment is determined by the accelerator model. There is a long-run unit elasticity between investment and output in all countries, but the long-run Žsemi-.elasticity of investment with respect to the Žlong-term. interest rate varies across countries. The stockbuilding equations are similar, except that the long-run relationship concerns the level of stocks and output. 2.4. Future expectations Some areas in which the models incorporate unobserved expectations of future values of endogenous variables are noted in the previous sections: expected future inflation influences current wage or price setting behaviour in all three models; a forward-looking definition of human wealth appears in MSG2 and MULTIMOD; the latter model employs an equivalent formulation of the real market value of the firm’s capital stock; in MSG2 Tobin’s q depends on the expected future path of firms’ profitability. The remaining areas of forward-looking behaviour concern the


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

determination of exchange rates and interest rates. In all cases exchange rates obey the uncovered interest parity or open arbitrage condition, in which the expected future appreciation of the currency is equal to the interest rate differential in its favour over the same holding period, a year or a quarter as appropriate. Similarly, an arbitrage condition determines the relation between short-term and long-term interest rates. All three models treat these unobserved future expectations as ‘rational’ or ‘model-consistent’ expectations. That is, in solving the model for the endogenous variable values over a simulation horizon, an internally consistent forward-looking solution is calculated, in which each period’s future expectations variables coincide with the model solution for the future period. A terminal condition or transversality condition that specifies the solution values and expectations at the end of the simulation horizon or beyond is then required. This may be based on analysis of the long-run equilibrium properties of the model and hence may require a relatively long solution period, to ensure that the model has reached an approximate equilibrium. In the absence of an explicit description of the equilibrium, or in a shorter solution period, terminal conditions that approximate a stable convergence to equilibrium are typically specified, by requiring constant growth rates or constant levels of forward-looking variables as appropriate. The MSG2 model uses the former approach; MULTIMOD and NIGEM the latter.

3. Simulation experiments 3.1. Experimental design We first describe the common policy framework in which our comparative simulation experiments are conducted. Since different models may give different simulation responses simply due to different assumptions about the underlying fiscal and monetary policy regimes, it is important that these should be standardized, so that the remaining differences genuinely represent differences in the models. This has not been accomplished in the model comparison conference style of project, Bryant et al. Ž1993. reporting, for example, that ‘when preparing their simulations, almost every model group departed in some way from the original specifications of the policy regimes’ Žp. 229.. To address this issue in the multicountry model comparison context is a leading objective of the present research. Three simulation experiments are undertaken, namely an increase in government expenditure, an increase in Žthe target level of. the money supply and an increase in the world price of oil; all are permanent, unanticipated changes. In each case the response of the model’s endogenous variables to the shock is estimated by comparing the results of two model solutions, one a ‘base’ run and the other a perturbed run in which the indicated exogenous variable is perturbed from its base-run values. Fiscal policy in all three models is subject to a reaction function or fiscal closure rule that imposes the government’s intertemporal budget constraint and so ensures

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


the sustainability of policy, in particular ruling out the possibility of an explosion of government debt. Although the budget constraint is silent on the question of which of the government’s income and expenditure variables should be modified in response to a perturbation, the closure rules that implement the constraint in the present models adjust direct taxation, although three different formulations of the target are employed. The MULTIMOD rule targets the debt stock, the NIGEM rule targets the budget deficit, expressed as a ratio to GDP, while the MSG2 rule implicitly targets debt interest payments, adjusting taxes to cover the interest burden of a changing debt stock by making interest payments a lump-sum tax on households. This rule implies that, following a permanent increase in the government expenditurerGDP ratio, the deficitrGDP ratio also rises in steady state, with the primary deficit Žspending net of interest payments less total tax receipts. eventually moving into surplus to prevent the explosive growth of government debt. The steady-state debtrGDP ratio is thus determined endogenously, whereas the MULTIMOD and NIGEM rules require target values of the relevant variable to be specified exogenously. This introduces a degree of arbitrariness when permanent shocks are implemented, as is the case here, and we avoid this by standardizing on the MSG2 rule, applying this in each country to which the shock is applied, except that this cannot be implemented in the NIGEM model of Japan. Church et al. Ž1996. undertake further analysis of the properties of the various rules and consider the sensitivity of simulation results to the different possibilities. In the second Brookings exercise ŽBryant et al., 1993. model proprietors were asked to switch off the closure rule for the first 10 years of any simulation. The delay in implementation of the rule appears to have been intended to avoid the contamination of simulations of, for example, a fiscal expansion by a countervailing fiscal contraction, and to increase the comparability of results with those models, typically backward-looking, that did not specify a fiscal closure rule. This restriction was criticized by Sims Ž1993., in his commentary on the exercise, primarily on the grounds of the time-inconsistency of such a policy regime, and in any event MSG2 and MULTIMOD appear not to have followed the instructions. The rule employed here operates continuously yet achieves a similar objective, as it operates relatively slowly to reduce the government budget deficit in the face of an expenditure increase. It serves more as a form of transversality condition on the government debt stock, preventing debt from exploding through accumulating interest payments in the longer term. This would seem to be the primary intention of those proposing the use of such closure rules in the context of models with forward-looking, model-consistent expectations, such as the three models under consideration here. The monetary policy regime adopted, in OECD countries, is that of medium-term monetary targeting, analogous to the treatment adopted in leading countries in MULTIMOD. National monetary authorities are assumed to target the value of the money stock over a number of years, the precise period being determined by the dynamics of money demand. Each of the three models employs a standard demand for money equation relating real money balances to the level of domestic aggregate activity, usually real GNP, and the nominal short-term interest rate.

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Although the dynamic specifications vary, the general form is log Ž MrP . t s a 0 q a 1 log Yt q a 2 RSt q a 3 log Ž MrP . ty1


where M is the narrow definition money supply, P is the GDP deflator, Y is GDP and RS the short-term interest rate. The long-run income elasticity, a 1rŽ1 y a 3 . in this example, would be expected to be unitary and the interpretation of the interest elasticity of money demand is therefore as the elasticity of substitution between money and the final good. The equation for the instrument of monetary policy, the short-term interest rate, is obtained by inverting the steady-state form of the money demand equation, which for Eq. Ž6. gives RSt s



wŽ 1 y a 3 . log Ž M ) rP . t y a 0 y a 1 log Yt x


where M U is the target value of the money stock. In Brookings terminology ŽBryant et al., 1993, pp. 12]13., this form of rule implies partial instrumental adjustment ŽPIA. when used in conjunction with a dynamic demand for money function. That is, the interest rate instrument adjusts so as to reduce partially, but not necessarily to eliminate completely, the deviations of the target variable from a desired time path, the speed of adjustment reflecting the dynamic specification. This is contrasted with full instrument adjustment ŽFIA., period-by-period, or, in British terminology, a Type II fix, which in dynamic models may face the problem of instrument instability. In the MSG2 model the money demand equation is static Ž a 3 s 0. so the above rule implies that PIA and FIA are equivalent in this case. Otherwise the rule provides a means of making the results from quarterly and annual models more easily comparable. Some sensitivity analysis with respect to the different coefficient values is presented in Sec. 3.5. This monetary policy setting does not apply to member countries Žas of 1992. of the European Exchange Rate Mechanism ŽERM., apart from Germany. Members are assumed to target their exchange rate against the deutschmark while Germany follows a medium-term money target in line with the other OECD countries, as above. In NIGEM and MSG2 this exchange rate targeting is assumed to be exact from period to period, whereas in MULTIMOD a non-linear function keeps exchange rates within a band of 1% around the existing rate. This difference has little impact on the simulations reported below. In general the monetary policy environment has been chosen for model comparison purposes and not as a description of current policy-making arrangements. The analysis of a range of policy regimes, possibly more realistic, is on our future research agenda. 3.2. Fiscal policy shock The first simulation is that of a permanent increase in government consumption expenditure of 1% of baseline GDP in each of the US, Japan, Germany and the UK in turn. The period of solution is the full length of the baseline supplied with

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


the model, namely: MSG2, 50 years; MULTIMOD, 65 years; and NIGEM, 19 years. However, the tabulated results are limited to a horizon of 15 years, which in all cases is sufficiently long for the models’ steady-state responses to emerge. Broadly similar simulation responses might be expected given the general characteristics of the three models described above. The standardized fiscal policy closure rule implies that a permanently higher level of government consumption spending generates a permanently higher level of government deficit and debt stock. A corollary is that the domestic country also generates a permanently higher level of current account deficit and overseas debt. There is crowding out of domestic private expenditure and lower long-run GDP, which is also consistent with a lower long-run capital stock. The impact of the fiscal expansion on other countries is clearer in the long run than in the transition phase. Lower domestic output implies reduced demand for imports which, when taken together with higher world interest rates and an appreciated exchange rate, means lower output for third-party countries. The size of this impact depends on the exposure of the country concerned to the country in which the shock originates. In the short run the size of the fiscal multiplier effect on GDP depends greatly on the response of interest rates and the size of the short-run interest elasticities of private expenditure components. The results for the US fiscal shock ŽTable 1. are broadly consistent with the Mundell]Fleming model. In each case there is significant crowding out of private sector expenditure through increased nominal and real interest rates and appreciation of the nominal and real exchange rate. All three models show initial positive multiplier effects on output, although in MULTIMOD this lasts for only 1 year compared with at least 5 years for the other two models. The first year impact multipliers are 0.77 in MULTIMOD, 1.13 in NIGEM and 0.33 in MSG2, thus the quantitative effects differ substantially across the models. The significantly lower impact multiplier for the MSG2 model appears to be due to two main influences. The first is the size of the increase in nominal and real interest rates and the second is the size of the response of net exports. The size of the jump in interest rates is locally determined by the interest rate elasticity of the demand for money, and the relatively large increase in nominal and real interest rates in the MSG2 model is explained by its relatively low interest elasticity. The private expenditure components of aggregate demand are in turn restrained by this interest rate increase. Private consumption and investment in the US model of MSG2 fall during the first year of the simulation, offsetting part of the initial expenditure increase. In comparison, private consumption in both NIGEM and MULTIMOD increase, by around 0.1 and 0.2% of base GDP. Private investment falls slightly in MULTIMOD, whereas in NIGEM, total investment also increases by some 0.2% of base GDP, primarily arising due to an increase in stockbuilding. The second factor relates to trade flows, where the response of net exports to the appreciation of the real exchange rate in MSG2 is substantially faster than in the other models. Both MSG2 and MULTIMOD experience similar appreciations of the US real exchange rate during the initial years of the simulation, but the fall in exports and rise in imports that result occur more slowly in MULTIMOD. In


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 1 Sustained increase in US Government Expenditure by 1% of GDP. Standardised fiscal and monetary rules.

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15





0.77 y0.04 y0.23 y0.05 y0.07

0.33 0.30 0.13 y0.12 y0.16

1.13 0.71 0.17 y0.31 y0.04

0.64 0.42 0.14 y0.16 y0.15

0.21 y0.29 y0.47 y0.49 y0.53

y0.01 y0.02 y0.15 y0.44 y0.60

0.09 y0.15 y0.60 y1.10 y0.92

y0.05 y0.17 y0.15 y0.02 y0.04

y0.03 0.00 y0.03 y0.11 y0.14

0.22 0.13 y0.08 y0.24 y0.14

0.39 1.07 0.96 0.21 0.30

0.03 y0.03 0.17 0.51 0.56

0.44 1.34 1.46 0.71 0.38

0.39 0.25 y0.14 y0.04 0.01

y0.27 0.03 0.12 0.05 0.02

0.45 0.26 0.02 y0.16 y0.01

0.26 0.38 0.29 0.04 0.08

0.74 0.67 0.77 0.95 0.93

0.33 0.43 0.35 0.09 0.08

0.33 0.28 0.15 0.06 0.07

0.81 0.86 0.91 0.93 0.91

0.27 0.20 0.14 0.09 0.12

3.14 2.79 2.39 2.05 1.96

5.05 4.26 4.03 3.35 2.56

0.70 0.21 y0.08 y0.24 y0.50

0.20 0.61 1.19 1.80 1.69

0.46 0.61 0.80 0.98 0.91

4.62 4.22 3.99 3.20 2.39

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 1 Ž Continued. MULTIMOD REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15



2.49 2.91 2.78 2.02 1.85

4.88 3.79 3.61 3.08 2.31

1.01 0.87 0.45 0.01 y0.11

y0.12 y0.29 y0.31 y0.33 y0.38

y0.41 y0.47 y0.52 y0.63 y0.70

y0.22 y0.34 y0.21 0.02 y0.01

0.70 0.81 0.86 0.84 0.86

0.80 0.88 0.95 1.05 1.10

0.68 0.56 0.27 0.07 0.05

0.09 0.47 0.17 y0.24 y0.02

0.00 0.00 0.00 0.00 0.00

y0.14 y0.33 y0.14 0.12 0.03


y0.40 y0.47 y0.53 y0.63 y0.71

Notes: See Appendix.

MSG2, exports fall by around 0.2% of base GDP and imports rise by around 0.4% of GDP during the first year, giving an overall worsening of the trade balance of some 0.6% of GDP. In MULTIMOD, the trade balance worsens by around 0.4% of GDP during the first year, not reaching 0.6% of GDP until year 5. This is because MSG2 models trade flows in domestic currency as direct proportions of domestic consumption and investment spending in any period. Therefore any change in relative prices passes through directly onto volume demands in any country, in addition to the passing on of domestic demand increases. In this case, domestic US demand for imports increases and overseas demand for US exports reacts quickly. The US real exchange rate appreciation in NIGEM is much less than that in the other models, producing smaller import and export responses; together with the consumption and investment effects, it is then not surprising that NIGEM yields the largest impact multiplier. Higher inflation tends to occur in all of the models along the simulation path, although in MSG2 inflation falls in the first year. This is due to the nature of wage formation in this model, discussed above, in which nominal wages are predetermined and therefore unchanged in the first year of the simulation. Given that output then rises more than employment as a result of the increase in demand,


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

labour productivity rises and marginal costs fall producing lower output prices in the first year. The response of wages along the simulation path, however, is to increase as future contract wages increase, anticipating future price increases and lower unemployment. Over the longer term the consequences of the government debt financing rule begin to show in the government deficit and current account to GDP ratios. In all cases the long-run effect of the rule is a higher deficit and higher government debt, together with a higher current account deficit and overseas debt for the domestic country, the size of both of these developments being larger in MULTIMOD than in the other models. The spillover effect of US fiscal expansion on the rest of the world depends in part on the relative importance of the US in the trade of the country concerned. Summary results for Japan, Germany and the UK are given, for each model in turn, in the upper sections of Tables 2]4. Overall the increase in US interest rates results in higher world interest rates, reflecting the size of the US economy. This expenditure-reducing effect is offset in some countries by the impact of the increase in US demand for imports and the relative price effect of a depreciation in the real exchange rate. The ranking of the impact spillover effects is the same for MULTIMOD and NIGEM but different for MSG2, which can be explained by the latter’s use of more up-to-date data. The trade weights in MSG2 are from UN Trade Statistics 1991, whereas the other two use the 1987 edition of the same publication. In ERM countries a monetary easing is required to maintain exchange rate parity. In MSG2, this results in the UK having the largest initial output rise. The 0.27% first-year increase in GDP is reduced to 0.11% if it is assumed that the UK is targeting the domestic money supply in the same way as non-ERM countries. The remaining difference in the initial output responses appears to be due to differences in trade shares. This issue is clearer for the monetary shock discussed below. Tables 5]7 contain the results of simulations of the same permanent increase in government expenditure Ž1% of GDP. in Japan, Germany and the UK, respectively. The main characteristics of its impact on the domestic economy are similar in nature to those for the US economy discussed above. There are, however, some quantitative differences which emerge between country shocks and models. In the case of the Japanese government expenditure increase, the GDP response is both smaller in impact Žall three models. and shorter in duration ŽMSG2 and NIGEM., than for the US. The two factors at work are the greater degree of openness of the Japanese economy and the greater flexibility of wages in Japan compared with the US. The first manifests itself as an appreciation in the nominal and real exchange rate. In both MSG2 and MULTIMOD a more substantial long-run appreciation in the exchange rate is required in Japan than was evident for the US. Consequently, the initial jump appreciation in the yen against all currencies is also larger. The corresponding appreciation in NIGEM is somewhat smaller than in the other two models but again is bigger than that in the US simulation. In MSG2 and NIGEM exports clearly fall by more in Japan than they do in the US, thus the negative trade balance effects are also stronger leading to faster crowding out of the

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 2 MULTIMOD results for other countries. Standardised fiscal and monetary rules

Sustained increase in US government expenditure by 1% of GDP GDP 1 3 5 10 15 PGNP 1 3 5 10 15 RS 1 3 5 10 15 ER 1 3 5 10 15 CArGDP 1 3 5 10 15 Sustained 5% increase in US monetary target GDP 1 3 5 10 15 PGNP 1 3 5 10 15




0.12 y0.07 y0.15 y0.11 y0.12

0.13 y0.05 y0.08 y0.07 y0.09

y0.17 0.06 0.80 0.98 y0.40

0.17 0.40 0.38 0.12 0.10

0.17 0.38 0.36 0.16 0.19

y0.12 y2.29 y3.83 y0.60 0.05

0.11 0.14 0.13 0.04 0.04

0.10 0.13 0.11 0.03 0.06

0.05 0.08 0.08 0.05 0.05

y3.99 y3.68 y3.32 y3.00 y2.91

y3.24 y2.90 y2.44 y1.97 y1.90

y3.41 y2.98 y2.38 y1.93 y1.88

0.17 0.62 1.11 1.20 1.41

y0.02 0.13 0.16 0.12 0.12

0.10 y0.18 y0.31 0.53 0.17

y0.02 y0.12 y0.07 0.00 y0.01

y0.08 y0.14 y0.03 0.02 0.00

y0.02 0.01 0.84 1.08 y0.28

y0.02 0.03 0.04 0.04 y0.03

y0.11 y0.12 0.04 0.05 y0.03

y0.25 y2.21 y3.53 y0.35 0.15


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 2 Ž Continued.

RS 1 3 5 10 15 ER 1 3 5 10 15 CArGDP 1 3 5 10 15




y0.04 y0.01 0.00 0.01 y0.01

y0.10 y0.07 y0.02 0.00 0.00

0.00 y0.01 0.00 0.00 0.00

5.77 4.63 4.57 4.66 4.75

5.94 4.82 4.81 4.95 5.04

6.01 4.85 4.75 4.95 5.05

0.55 0.54 1.53 0.90 0.66

0.13 y0.05 0.08 0.08 0.03

0.17 y0.53 y0.71 0.03 y0.29

Notes: See Appendix.

government expenditure increase. In NIGEM, it is not only this trade balance effect which helps to attenuate the positive output effect. A large fall in business investment causes total investment to fall from the onset of the simulation. A combination of these two effects leads to the quicker crowding out of government expenditure in this model. It is again evident that government and overseas debt grow somewhat less in the NIGEM model. The second factor is the much faster response of wages and prices in all three models of Japan, discussed above. This is most extreme in the MSG2 model where the wage rate jumps after one period of the simulation to wipe out almost completely the initial gains in output. The disequilibrium created in the labour market in the first year of the simulation is expected by wage setters to have disappeared in subsequent years and so unemployment is lower for 1 year only, returning to its base value in year 2. Despite the anticipated rise in prices which one would expect to be part of the process of crowding out of private sector spending, inflation falls initially in MSG2. The cause is the fall in first year marginal cost and therefore prices, given a predetermined wage rate, as in the US results. In this case the initial fall in prices is even larger. The real interest rate rises less in Japan than in the US, reflecting faster price adjustment. However, the response of the nominal interest rate in MSG2 is again out of line with the other two models, again due to the interest elasticity of money demand. The leading European economies are more open than either the US or Japan. From figures employed in the MSG2 model, Germany and the UK import approxi-

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 3 MSG2 results for other countries. Standardised fiscal and monetary rules

Sustained increase in US government expenditure by 1% of GDP GDP 1 3 5 10 15 PGNP 1 3 5 10 15 RS 1 3 5 10 15 ER 1 3 5 10 15 CArGDP 1 3 5 10 15 Sustained 5% increase in US monetary target GDP 1 3 5 10 15 PGNP 1 3 5 10 15




0.02 y0.04 y0.07 y0.12 y0.14

0.09 y0.11 y0.22 y0.37 y0.43

0.27 y0.05 y0.20 y0.22 y0.22

0.13 0.44 0.53 0.64 0.64

0.10 0.46 0.63 0.85 0.91

0.20 0.83 1.20 1.42 1.51

0.10 0.52 0.62 0.74 0.74

0.26 0.51 0.61 0.72 0.72

0.25 0.50 0.60 0.72 0.72

y5.81 y4.93 y4.64 y3.70 y2.69

y7.09 y6.36 y6.06 y5.02 y3.87

y7.06 y6.31 y6.00 y4.94 y3.79

0.54 0.53 0.54 0.55 0.53

0.52 0.44 0.40 0.36 0.33

0.58 0.53 0.53 0.62 0.69

y0.18 0.01 0.01 y0.01 0.00

y0.31 0.05 0.03 y0.09 y0.06

y0.64 0.12 0.18 y0.09 0.01

y0.19 y0.30 0.05 0.06 y0.01

y0.21 y0.35 y0.02 0.14 0.05

y0.37 y0.71 y0.34 0.17 y0.06


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 3 Ž Continued.

RS 1 3 5 10 15 ER 1 3 5 10 15 CArGDP 1 3 5 10 15




y0.48 y0.47 0.06 0.08 y0.02

y0.84 y0.48 0.01 0.09 y0.01

y0.85 y0.49 0.00 0.09 y0.01

7.30 4.33 4.08 4.74 4.81

6.85 4.16 3.97 4.75 4.76

6.80 4.13 3.95 4.75 4.75

y0.43 y0.03 0.09 0.02 0.00

y0.44 y0.03 0.07 y0.02 y0.03

y0.52 y0.05 0.09 y0.02 y0.01

Notes: See Appendix.

mately 16% of their consumption goods and between 30 and 40% of their investment goods whereas the figures for the US are 7 and 16% and for Japan are 9 and 3%, respectively. Consequently, appreciation of the exchange rate offsets the short-term expenditure increasing effect of the fiscal expansion. The appreciation in Germany shown in Table 6 is, however, not as large as in Japan, which is due to the lower degree of wage flexibility in Germany. The longer term impact of the sluggishness of the German labour market is especially noticeable in the MSG2 model, where near-hysteresis in the unemployment rate results in higher GDP and lower unemployment throughout the simulation period shown in the Tables. This is also evident to a reduced degree in the other models. Positive output effects are sustained for 27 years in MSG2 and 6 years in NIGEM. These can be contrasted with the output responses for Japan which are sustained for 7 and 4 years, respectively. In MULTIMOD, the distinction is far less clear, although output is slightly slower in returning to base, with virtually no overshooting. In both NIGEM and MULTIMOD the negative trade balance effects which result from the shock are larger for Germany than those for Japan, while in MSG2 they are comparable. Thus, the sustained positive output effects in Germany are due to differences in the behaviour of other interest-sensitive components of aggregate demand. In MSG2 and MULTIMOD, consumption falls below base more quickly in Japan than in Germany, while in NIGEM, Japanese consumption lies below base from the onset of the simulation, immediately offsetting part of the government expenditure increase. In all three models Japanese investment falls below base, remaining so in MSG2 for the whole simulation period, whereas in MSG2 and NIGEM,

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 4 NIGEM results for other countries. Standardised fiscal and monetary rules

Sustained increase in US government expenditure by 1% of GDP GDP 1 3 5 10 15 PGNP 1 3 5 10 15 RS 1 3 5 10 15 ER 1 3 5 10 15 CArGDP 1 3 5 10 15 Sustained 5% increase in US monetary target GDP 1 3 5 10 15 PGNP 1 3 5 10 15 RS 1 3 5 10 15




0.19 0.19 0.00 y0.12 y0.03

0.20 0.26 0.18 y0.20 y0.12

0.09 0.07 0.00 y0.18 y0.08

y0.02 0.37 0.60 0.25 y0.02

y0.03 0.23 0.45 0.48 0.06

0.01 0.20 0.37 0.19 y0.20

0.06 0.15 0.16 0.04 y0.01

0.06 0.14 0.17 0.08 y0.01

0.06 0.14 0.17 0.08 y0.01

y1.06 y0.50 y0.08 0.37 0.63

y0.80 y0.23 0.20 0.49 0.70

y0.80 y0.23 0.20 0.49 0.70

0.15 0.22 0.12 0.00 y0.01

0.07 0.15 0.11 y0.04 y0.02

0.09 0.22 0.20 0.02 y0.04

y0.22 y0.13 0.12 0.09 y0.02

y0.33 y0.02 0.13 0.29 0.06

y0.19 y0.17 y0.03 0.14 y0.06

0.01 y0.44 y0.52 0.00 0.09

0.11 y0.39 y0.49 y0.18 0.21

y0.02 y0.52 y0.82 y0.42 0.03

y0.07 y0.15 y0.10 0.02 0.02

y0.12 y0.12 y0.10 0.03 0.08

y0.12 y0.12 y0.10 0.03 0.08


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 4 Ž Continued. Japan ER 1 3 5 10 15 CArGDP 1 3 5 10 15



5.99 4.99 4.65 4.73 4.81

6.01 5.03 4.64 4.79 4.69

6.01 5.03 4.64 4.79 4.69

y0.20 y0.07 0.09 0.09 0.01

y0.03 0.08 0.10 0.05 y0.04

y0.12 y0.07 0.04 0.06 0.00

Notes: See Appendix.

German investment is above base until years 24 and 4, respectively, strengthening the GDP response. In MULTIMOD, German investment falls slightly below base during the first 3 years of the simulation, a fall much smaller in magnitude than that for Japan. The impact of exchange rate parity with Germany is evident in the results of the UK fiscal expansion shown in Table 7. In MSG2 and MULTIMOD, the impact effects on output are larger than for any of the other countries. In NIGEM and MULTIMOD, the positive effects are more sustained. These effects can be attributed to the associated accommodation in monetary policy. In order that sterling not appreciate against the deutschmark, it is assumed in all cases that the short-term interest rate is reduced. Thus, the interest-sensitive components of aggregate demand react differently in this simulation, as they are not crowded out by higher interest rates. In MSG2 and MULTIMOD, UK consumption and investment increase by more than in any other country, with the positive effects in MULTIMOD lasting for around 13 years. In NIGEM, the sustained GDP response is due to investment rather than consumption effects. The interest rate responses do not stimulate consumption and investment any more than in any other country, although the negative investment effects which occur from year 4 onwards are much smaller in magnitude, keeping GDP above base for longer. Alternative results for the UK fiscal expansion, not given in Tables, where it is assumed that the UK targets the money supply, rather than the DM exchange rate, are much more similar to those for Germany in Table 6. Further analysis of the impact of alternative monetary policy assumptions for all, especially European, economies is clearly warranted. By the end of the simulation period MSG2 and NIGEM see the whole of the government expenditure increase offset by lower consumption and investment, with the former offsetting almost all of the shock. In MULTIMOD the government expenditure increase is offset by a combination of lower consumption and a

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 5 Sustained increase in Japan government expenditure by 1% of GDP. Standardised fiscal and monetary rules Žexcept for NIGEM.

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15




0.66 y0.06 y0.19 y0.06 y0.06

0.24 0.04 0.02 y0.01 y0.03

0.73 0.17 y0.06 y0.13 y0.07

0.18 y0.19 y0.29 y0.30 y0.31

0.27 0.18 0.11 y0.08 y0.25

y0.02 y0.05 y0.26 y0.80 y1.10

y0.06 y0.24 y0.21 y0.02 0.00

y0.06 y0.08 y0.09 y0.09 y0.10

y0.03 y0.25 y0.39 y0.30 y0.03

0.30 0.90 0.81 0.36 0.37

y0.21 y0.02 0.00 0.06 0.13

0.23 1.00 1.03 0.50 0.09

0.30 0.22 y0.10 y0.02 0.00

y0.54 y0.02 0.00 0.02 0.03

0.23 0.35 y0.07 y0.10 y0.06

0.21 0.34 0.25 0.07 0.06

0.44 0.49 0.46 0.44 0.46

0.23 0.26 0.21 0.09 0.01

0.28 0.24 0.15 0.06 0.06

0.47 0.45 0.45 0.46 0.47

0.18 0.14 0.10 0.01 y0.03

6.92 6.53 5.99 5.50 5.35

6.05 5.19 4.71 3.93 3.13

1.29 0.87 0.50 0.12 0.11


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 5 Ž Continued. MULTIMOD REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15



3.77 4.50 4.72 4.74 5.08

6.06 5.27 4.70 3.92 3.16

1.43 1.56 1.11 0.28 0.02

y0.73 y1.45 y1.71 y2.19 y2.56

y0.84 y0.85 y0.84 y0.86 y0.87

y0.15 y0.28 y0.22 y0.05 y0.01

0.80 0.88 0.88 0.81 0.79

0.94 0.93 0.93 0.91 0.90

0.93 0.60 0.64 0.52 0.23

y0.02 0.39 0.10 y0.21 y0.08

0.00 0.00 0.00 0.00 0.00

0.03 0.40 0.15 y0.11 y0.11

Notes: See Appendix.

worsening of the trade balance. The weights attached to these two effects vary across countries. In the UK the worsening of the trade balance offsets most of the expenditure increase, whereas in the US, the offsetting effect is split more evenly between the two. 3.3. Monetary shock This simulation is of a permanent, unanticipated increase of 5% in the target level of the money supply. This shock is again applied to each of four countries in turn and the results are presented in Tables 8]11. While the target level of the money supply increases by the same amount in each model, there is a difference between the models in the response of the measured nominal money supply which, as discussed above, depends on the money demand function. In MULTIMOD and NIGEM the short-run dynamics of money demand result in the response of the money supply being damped. For the US shock, for example, in MULTIMOD the long-run change in the money supply is reached in 3 years but in NIGEM full adjustment takes 14 years, although over 90% is accomplished after 3 years. In

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 6 Sustained increase in German government expenditure by 1% of GDP. Standardised fiscal and monetary rules MULTIMOD GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15



0.47 0.04 0.01 0.04 0.00

0.19 0.22 0.27 0.25 0.17

0.95 0.54 0.16 y0.22 y0.10

0.12 y0.12 y0.23 y0.35 y0.43

0.14 0.19 0.18 0.03 y0.12

0.12 0.02 y0.27 y0.94 y1.08

y0.03 y0.03 y0.01 0.02 0.00

0.01 0.07 0.10 0.07 0.04

0.31 0.11 y0.07 y0.23 y0.17

0.15 0.32 0.30 0.15 0.14

0.03 y0.17 y0.29 y0.27 y0.15

0.11 0.57 0.80 0.44 y0.03

0.15 0.06 y0.03 y0.03 y0.01

y0.17 y0.11 y0.04 0.03 0.02

0.11 0.21 0.09 y0.13 y0.06

0.20 0.12 0.07 0.03 0.05

0.47 0.26 0.15 0.12 0.16

0.28 0.27 0.23 0.05 y0.03

0.13 0.08 0.04 0.04 0.03

0.20 0.14 0.13 0.15 0.17

0.18 0.13 0.07 y0.01 y0.02

1.78 1.52 1.37 1.24 1.09

3.83 3.08 2.78 2.48 2.19

1.22 0.73 0.31 y0.19 y0.14


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 6 Ž Continued. MULTIMOD REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15



0.79 1.32 1.57 1.31 0.78

1.64 1.81 1.83 1.52 1.32

0.27 0.58 0.72 0.36 y0.05

y0.26 y0.29 y0.24 y0.29 y0.31

y0.82 y0.82 y0.83 y0.87 y0.93

y0.36 y0.43 y0.37 y0.08 0.05

0.88 0.89 0.87 0.91 0.94

0.87 0.84 0.81 0.80 0.81

0.93 0.54 0.36 0.09 y0.12

y0.06 y0.08 y0.11 y0.04 0.02

0.00 0.00 0.00 0.00 0.00

y0.01 0.06 0.07 y0.02 y0.04

Notes: See Appendix.

MSG2 the money supply jumps immediately to its long-run level. Results for the other countries’ money target shocks are broadly similar. The long-run impact of this shock, in all models, is expected to be an increase in all nominal quantities of 5%, with variables measured in real terms unchanged from their base values. This long-run static homogeneity can be observed in the results in Tables 8]11 and is a design feature of all three models. The major differences between the models are found in their short-run responses, although in all cases the general features of the dynamic Mundell]Fleming model emerge. A domestic monetary expansion does not result in an immediate rise in prices due to price and wage stickiness, which varies across countries and models, as discussed in Sec. 2.2. We begin with the results for the US monetary shock in Table 8. The response of the short-term interest rate is the key transmission mechanism for a monetary shock, and the initial response can be ranked in order of the change in the money supply, with a larger response in MSG2 than in MULTIMOD and NIGEM due to the immediate adjustment of the money supply to its new long-run level. The associated jump depreciation of the nominal and real exchange rate is not ranked in the same way, given the differing degrees of price stickiness in the models

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 7 Sustained increase in UK government expenditure by 1% of GDP. Standardised fiscal and monetary rules MULTIMOD GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15



0.90 0.87 0.84 1.11 y0.17

1.30 0.50 0.07 0.05 0.10

0.64 0.60 0.38 y0.21 y0.24

0.21 0.12 0.10 0.36 y0.14

0.68 0.35 0.14 y0.08 y0.26

0.03 0.02 y0.24 y1.08 y1.36

0.09 0.17 0.20 0.27 y0.10

0.37 0.29 0.22 0.16 0.11

0.01 0.03 0.01 y0.08 y0.10

0.64 0.87 0.25 1.49 3.00

0.67 2.04 2.68 2.14 1.51

0.07 0.57 1.10 0.71 y0.36

0.64 0.26 y0.74 0.91 y0.38

0.56 0.55 0.19 y0.16 y0.08

0.07 0.30 0.23 y0.22 y0.18

0.00 0.00 0.00 0.01 0.11

y0.03 0.00 0.07 0.11 0.10

0.01 0.02 0.03 0.02 y0.01

0.00 0.00 0.00 0.05 0.11

0.06 0.09 0.10 0.10 0.10

0.02 0.02 0.02 y0.01 y0.03

0.13 0.14 0.16 0.17 0.17

0.82 0.89 0.93 0.81 0.65

0.17 0.15 0.13 0.09 0.13


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 7 Ž Continued. MULTIMOD REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15



0.49 0.60 0.43 1.16 2.23

1.04 2.47 3.12 2.50 1.78

0.09 0.50 0.92 0.59 y0.26

y0.37 y0.37 y0.52 y0.32 y0.50

y0.63 y0.87 y1.03 y1.16 y1.26

y0.36 y0.37 y0.30 y0.05 0.11

0.62 0.53 0.59 0.49 0.74

0.85 1.04 1.10 1.22 1.41

0.89 0.57 0.45 0.30 0.12

0.63 0.99 0.58 1.82 2.83

0.00 0.00 0.00 0.00 0.00

0.28 1.00 1.38 0.49 y0.52

Notes: See Appendix.

concerned. The first year depreciation in NIGEM is larger than in MULTIMOD, which is associated with greater price stickiness in the NIGEM model of the US economy. In all three models the nominal exchange rate initially overshoots its expected long-run depreciation of 5%. Results for the US monetary shock are broadly consistent with the Mundell]Fleming model, as all three models show initial positive multiplier effects on output which last for 13 years in NIGEM compared with 4 and 5 years in MSG2 and MULTIMOD. These domestic real effects have both domestic and overseas consequences. Lower real interest rates persist in all three models for at least 10 years, encouraging initially higher consumption and investment spending and improving the trade balance position. In NIGEM the sluggishness of trade volumes in responding to the substantial gain in domestic competitiveness and the slow response of consumption and investment to the lower interest rates mean that the positive effect on output takes 5 years to reach a maximum, thus explaining the relatively low impact multiplier in this model. In MULTIMOD and MSG2 the impact is more immediate and the associated jump in inflation ensures that it has unwound by the time 5 years have elapsed. In MSG2 consumption, investment and the trade balance improve immediately, as volume demands respond quickly to the

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 8 Sustained 5% increase in US monetary target. Standardised fiscal and monetary rules

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15





0.79 0.63 y0.25 0.07 0.03

2.12 0.38 y0.25 y0.07 0.03

0.38 1.08 1.09 0.41 y0.07

1.54 0.48 y0.10 0.04 0.02

0.24 0.32 y0.05 0.03 0.02

1.37 0.32 y0.10 y0.03 0.04

0.12 0.43 0.60 0.27 y0.10

0.32 0.15 y0.11 0.03 0.00

0.50 0.08 y0.06 y0.02 0.01

0.17 0.52 0.46 0.16 y0.02

1.23 5.92 5.92 4.80 4.99

1.23 3.80 4.85 4.66 4.49

0.01 1.64 2.96 4.95 5.13

1.23 1.57 0.40 y0.13 0.04

1.66 1.09 0.38 y0.10 0.00

0.01 0.83 0.63 0.21 y0.03

y1.20 y0.09 0.26 y0.06 0.00

y2.66 y0.73 0.20 0.13 y0.02

y0.90 y0.45 y0.19 0.07 0.01

y0.27 0.14 0.13 y0.04 0.00

y0.37 0.07 0.15 0.01 0.00

y0.20 y0.05 0.02 0.01 y0.03

y5.63 y4.34 y4.46 y4.84 y4.77

y7.80 y4.47 y4.16 y4.75 y4.82

y5.77 y4.97 y4.74 y4.79 y4.62

0.91 3.29 4.71 4.95 4.47

y1.26 y0.46 0.05 0.02 y0.02

y6.29 y4.63 y4.33 y4.76 y4.87


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 8 Ž Continued.

REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15




y3.27 0.17 1.12 y0.04 y0.08

y6.13 0.06 1.10 0.28 0.15

y5.03 y2.34 y1.08 y0.28 y0.12

y0.08 0.21 0.07 0.04 0.04

0.25 0.07 0.01 0.01 0.02

y0.11 0.00 y0.11 y0.15 y0.02

y0.12 y0.19 y0.06 y0.09 y0.08

y0.21 0.09 0.11 y0.01 y0.02

y0.10 y0.10 0.16 0.15 0.04

2.27 5.40 6.11 4.83 5.06

5.00 5.00 5.00 5.00 5.00

2.09 4.63 4.62 4.80 5.01


0.14 0.05 0.01 0.01 0.02

Notes: See Appendix.

change in the terms of trade. Any sluggishness in the pass-through to trade prices of changes in the nominal exchange rate is much less important in delaying the volume response than the slow response of trade volumes through the traditional trade equations of the other models. The net effect in MSG2 is that the current account improves immediately, positive volume changes more than offsetting negative revaluation effects. This is not true for the other two models. In MSG2 the positive effect on US domestic demand is amplified by an increase in private consumption of around 1.4% of base GDP. In comparison, private consumption in NIGEM and MULTIMOD increases by much less, approximately 0.6 and 0.55% of base GDP. As both the trade balance improvement and the increase in private investment that result are not radically different from the responses in NIGEM and MULTIMOD, the relatively large effect on output in the US model of MSG2 is primarily due to consumption effects. The spillover effects of the US money shock differ widely across country models, as can be seen from the lower sections of Tables 2]4. In general the impact on real quantities in any country depends on the relative importance of the appreciation of the exchange rate against the US dollar in reducing expenditure and the offsetting effect on exports of the expansion in US demand. Lower domestic interest rates in

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 9 Sustained 5% increase in Japan monetary target. Standardised fiscal and monetary rules Žexcept for NIGEM. MULTIMOD GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15



0.91 0.59 y0.27 0.10 0.03

1.77 0.04 0.03 0.02 0.01

0.68 1.25 0.68 0.09 y0.06

0.25 0.27 y0.02 0.04 0.04

0.20 y0.02 y0.01 y0.01 y0.01

0.03 0.27 0.64 0.56 0.36

0.49 0.25 y0.21 0.05 0.01

0.50 0.00 0.00 0.00 0.00

0.21 0.56 0.51 y0.07 y0.33

1.15 4.43 5.87 4.81 4.99

1.15 4.93 4.97 4.97 4.98

y0.07 1.81 3.83 5.34 5.53

1.15 1.58 0.40 y0.12 0.03

1.60 0.05 0.01 0.00 0.00

y0.07 1.18 0.83 0.16 y0.01

y1.24 y0.10 0.26 y0.05 0.00

y4.01 y0.06 0.01 0.00 0.00

y1.26 y0.45 y0.15 0.07 0.10

y0.29 0.14 0.13 y0.04 0.00

y0.43 0.00 0.00 0.00 0.00

y0.23 y0.05 0.03 0.09 0.08

y5.67 y4.12 y4.22 y4.84 y4.70

y8.98 y5.08 y5.00 y4.96 y4.97

y6.83 y5.31 y4.91 y4.97 y5.28


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 9 Ž Continued.

REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15




y2.29 0.03 0.80 y0.50 0.03

y8.13 y0.22 y0.12 y0.01 y0.02

y6.13 y2.91 y0.82 0.24 0.03

y0.66 y0.91 y1.31 y0.98 y0.85

0.51 0.01 0.01 0.00 0.00

0.34 0.32 y0.16 y0.19 y0.06

y0.12 y0.09 0.06 y0.01 0.00

0.35 0.00 0.00 0.00 0.00

y0.15 y0.90 y0.97 0.00 0.44

2.11 5.40 6.18 4.84 5.05

5.00 5.00 5.00 5.00 5.00

2.04 5.02 5.98 5.43 5.07

Notes: See Appendix.

the short run due to the US monetary expansion also provide an expenditure expanding effect in all countries. The resolution of these effects in MSG2 is for the appreciation in the exchange rate to dominate and a reduction in output in all other countries except Canada to occur. In general the size of the negative effect depends on the exposure of the economy concerned to the US economy and the impact of ERM membership. In Canada, by contrast, the spillover of the increase in US demand is more important resulting in an initial positive net output effect. The speed with which these real effects unwind depends primarily on the degree of price stickiness in the economy. In the ERM countries a degree of monetary tightening is required to maintain parity with the deutschmark as a result of the greater openness of the German economy and consequent greater appreciation. In MSG2 this results in the negative spillover effect being greatest in the UK. The initial output loss of 0.64% is reduced to 0.45% if the UK is assumed to be outside the ERM, targeting money in the same way as non-ERM countries, which is still a greater first-year loss than the 0.31% in Germany. The reason is that, according to the trade weights employed in MSG2, the UK gets more of its imports from the US than does Germany and the US more from Germany than the UK. Thus the positive US income effect is

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 10 Sustained 5% increase in German monetary target. Standardised fiscal and monetary rules MULTIMOD GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15



1.11 0.45 y0.32 0.11 0.04

2.34 0.70 0.50 0.38 0.34

0.72 1.48 1.54 0.71 y0.02

0.25 0.26 y0.01 0.08 0.04

0.70 0.19 0.10 0.14 0.15

0.12 0.70 1.14 0.88 0.01

0.44 0.22 0.14 0.07 0.01

0.50 0.10 0.02 0.03 0.04

0.31 0.82 0.85 0.39 0.00

1.32 4.55 5.76 4.80 4.96

1.57 4.23 4.64 4.79 4.78

y0.14 1.10 2.44 4.99 5.45

1.32 1.47 0.31 y0.10 0.03

1.85 0.71 0.13 y0.02 0.01

y0.14 0.66 0.68 0.33 y0.02

y1.12 y0.11 0.19 y0.08 0.00

y2.08 y0.40 y0.08 0.00 y0.06

y1.12 y0.64 y0.30 0.13 0.10

y0.28 0.09 0.08 y0.05 y0.01

y0.36 y0.07 y0.02 y0.05 y0.05

y0.28 y0.07 0.05 0.09 0.04

y5.78 y4.50 y4.57 y4.97 y4.85

y8.76 y5.90 y5.45 y5.32 y5.25

y6.75 y5.09 y4.35 y4.56 y5.12


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 10 Ž Continued.

REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15




y1.08 0.39 0.49 0.33 y0.10

y2.85 y0.87 y1.03 y0.58 y0.39

y1.06 y0.22 0.49 1.19 0.65

y0.08 0.07 0.05 0.05 y0.01

0.84 0.24 0.23 0.17 0.15

0.06 y0.05 y0.31 y0.38 y0.07

y0.21 y0.14 y0.02 y0.11 y0.06

y0.63 y0.24 y0.19 y0.15 y0.12

y0.03 y0.07 0.09 0.35 0.08

2.35 5.32 5.97 4.84 5.09

5.00 5.00 5.00 5.00 5.00

0.45 2.04 3.18 4.54 4.86

Notes: See Appendix.

greater for Germany than the UK, and this is reinforced by the negative relative price effect being smaller for Germany than the UK. In MULTIMOD and NIGEM this reasoning is reversed and Germany suffers a larger output-reducing relative price effect. These differences between the models are due to the parameters measuring the weight of countries in each others’ trade. As with the fiscal shock, the use of more up-to-date data in MSG2 distinguishes the behaviour in this model from that in the other two. Over time the spillover output effect disappears as prices adjust. In MSG2 this happens fastest in Japan where prices are only sticky for one period. It is most prolonged in Germany and the rest of the EMS where disequilibrium in the labour market is assumed to be most persistent. These differences are also reflected prominently in the results of money target increases in the three other countries shown in Tables 9]11. In order to implement a money target increase in the UK the monetary policy regime is altered to one of money targeting rather than targeting the deutschmark. The importance of the openness of the economy concerned and the differing degrees of wage and price stickiness can be gauged from the size of the initial output effect. The largest impact for all three models is in Germany and the UK

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 11 Sustained 5% increase in UK monetary target. Standardised fiscal and monetary rules

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15




0.40 1.46 1.02 1.02 y0.02

2.28 0.21 y0.26 y0.03 0.01

0.66 1.24 0.82 y0.05 y0.11

y0.09 0.45 0.31 0.40 y0.05

0.61 0.03 y0.08 0.00 0.00

0.23 1.07 1.09 0.04 y0.25

0.37 0.43 0.17 0.18 y0.08

0.30 0.02 y0.03 y0.01 0.00

0.15 0.23 0.13 y0.03 0.02

1.41 3.21 3.74 3.81 5.14

1.55 3.44 5.27 5.03 4.98

y0.17 2.33 4.32 5.38 4.90

1.14 1.15 y0.35 0.49 y0.11

2.19 0.90 0.15 y0.04 0.00

y0.18 1.26 0.89 y0.04 y0.08

y0.89 y0.26 y0.16 y0.19 0.07

y2.37 y0.42 0.09 0.04 0.00

y1.60 y0.55 0.04 0.16 y0.03

y0.40 y0.21 y0.28 0.01 0.01

y0.35 y0.11 0.05 0.01 0.00

y0.16 0.07 0.12 0.00 y0.06

y7.53 y6.28 y6.06 y4.88 y4.95

y8.41 y4.94 y4.47 y4.87 y4.94

y6.68 y4.53 y4.07 y5.00 y5.24


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 11 Ž Continued.

REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15




y3.89 y2.21 y1.88 y1.19 y0.39

y6.93 y0.35 0.76 0.13 0.04

y5.33 y1.79 0.24 0.41 y0.10

y1.20 0.02 0.16 0.23 0.10

0.67 0.03 y0.07 y0.01 0.00

y0.38 y0.16 y0.28 y0.06 0.04

0.14 y0.23 y0.18 y0.20 y0.05

0.06 0.11 0.04 y0.03 y0.03

y0.30 y0.35 0.21 0.35 0.07

3.02 4.82 5.13 5.18 5.43

5.00 5.00 5.00 5.00 5.00

1.73 5.41 6.27 5.05 4.88

Notes: See Appendix.

followed by Japan and the US. MULTIMOD and NIGEM put Japan ahead of the US, in contrast to MSG2. As with the US simulation, in NIGEM the beneficial output effect for the other three countries takes 3]5 years to reach a peak as a consequence of the slow adjustment of the interest-sensitive components of demand and of trade volumes to the gain in competitiveness. In MSG2 the positive effect on Japanese, German and UK domestic demand is amplified by a large improvement in the trade balance relative to the other models. This differs from the US simulation where relatively large consumption effects led to the large increase in output. The magnitude of the consumption effects in Germany and the UK are comparable with those in the other models, whereas the Japanese consumption response is much smaller, contributing towards the smaller impact effect on output in the Japanese simulation. In MSG2 the fast adjustment of the Japanese labour market is reflected in all of the results. Virtually all adjustment is complete by the second year of the simulation with the real exchange rate nearly back to its base value after the first year. This stands in contrast with the results for Germany where the output effects in MSG2 and NIGEM are sustained for at least 10 years due to the sluggishness of the German labour market. In NIGEM the relatively large positive effect on German GDP comes from a relatively large

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Fig. 1. Response of wages in MSG2.

increase in stockbuilding. The positive consumption and investment effects which follow the shock are sustained in the German simulation, in turn sustaining the output response. This differs from the MSG2 model where the sluggishness of the trade volumes in adjusting to the shock play a more important role in sustaining output levels. For these two models this simulation well illustrates the importance of crosscountry differences in the modelling of wages and prices, discussed in Sec. 2.2. The response of wages in MSG2 and NIGEM is shown in Figs. 1 and 2, respectively. In MSG2 wages in Japan reach the long-run level quickly. Wages in the US and the UK overshoot the steady state for a period while wages in Germany are extremely sluggish and deviate from the steady state over the full 50-year period. A corollary of these results is that employment and unemployment are affected by the nominal shock in Japan for only 1 year, whereas in Germany employment is 0.4% above base after 25 years. The NIGEM model ŽFig. 2. shows a more protracted approach to the steady state in general, although in this case Germany does converge too. The absence of the forward-looking component in the Japanese and German wage equations is reflected in their slower approach to the steady state, but this is a pairing that is unexpected and it is more difficult to relate the overall dynamic responses in NIGEM to the empirical evidence discussed above. In MULTIMOD the UK output response is an outlier. Positive output effects persist for 15 years compared to 5 years for the other three countries, with consumption, investment and trade balance effects all contributing to this response. Differences between the results for the US, Germany and Japan are less obvious in this model. The relative sizes of the overseas sector in demand for domestic goods and import supply provide the main sources of any differences that


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Fig. 2. Response of wages in NIGEM.

do arise, these results highlighting the importance of MULTIMOD’s common specification of wage and price determination, discussed in Sec. 2.2. Spillover effects from the three non-US money target shocks are generally very small. This reflects the well-known ability of flexible exchange rates to insulate countries from overseas monetary shocks. 3.4. World oil price shock We consider a permanent increase in the world oil price of 25%. As this simulation represents a supply-side shock, we expect to observe lower output associated with an increase in the price level. The initial effect of an increase in price of an imported input is to lower the income received by domestic factors of production and therefore the magnitude of the negative output effect reflects the relative importance of oil within the production process. The higher import prices which result following the oil price shock are passed on to domestic prices, with the speed and extent of adjustment differing between countries and across models. The simulation results are presented model-by-model in Tables 12]14. They show no obvious pattern across different countries and suggest that disagreement exists between models as to whether the effects diminish or increase over time. Results for the UK in MULTIMOD are not reported due to a problem with the UK oil imports equation. Failure of the identity to hold exacerbates the UK’s terms of trade response so that the UK appears to be exporting more oil than it is actually producing. Both NIGEM and MSG2 illustrate the adverse effect on output for all four countries of study, whereas in MULTIMOD German GDP lies slightly above base during the first year of the simulation. The impact effect on GDP

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 12 Sustained increase in world oil price by 25%. Standardised fiscal and monetary rules MULTIMOD

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15




y0.10 y0.12 0.02 y0.03 y0.02

y0.10 y0.09 y0.02 y0.09 y0.08

0.02 y0.06 y0.01 y0.08 y0.07

y0.09 y0.13 y0.08 y0.12 y0.14

y0.12 y0.15 y0.16 y0.25 y0.30

y0.06 y0.12 y0.13 y0.21 y0.29

y0.07 y0.05 0.00 y0.03 y0.03

y0.10 y0.11 y0.06 y0.09 y0.07

y0.06 y0.07 y0.04 y0.07 y0.08

0.44 0.08 y0.16 0.00 0.00

0.13 y0.04 y0.13 y0.04 y0.12

0.14 0.05 y0.03 0.05 y0.01

0.44 y0.18 y0.07 0.03 0.00

0.13 y0.10 y0.02 0.00 y0.01

0.14 y0.07 y0.03 0.01 y0.02

0.21 0.05 y0.03 0.02 0.02

0.17 0.05 0.01 0.03 0.01

0.13 0.05 0.01 0.03 0.04

0.07 y0.01 y0.01 0.02 0.01

0.07 0.02 0.02 0.02 0.01

0.06 0.02 0.01 0.04 0.02


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 12 Ž Continued. MULTIMOD US ER 1 3 5 10 15 REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15



0.61 0.49 0.49 0.62 0.69

y1.38 y1.36 y1.41 y1.66 y1.73

y0.73 y0.64 y0.67 y0.81 y0.94

0.36 0.33 0.35 0.32 0.24

y0.59 y0.61 y0.62 y0.93 y1.22

y0.07 0.10 0.21 0.04 y0.35

y0.14 y0.11 y0.06 y0.04 y0.01

y2.09 y1.35 y1.07 y0.93 y0.98

y0.08 y0.02 0.01 0.02 0.00

0.06 0.06 0.03 0.03 0.03

0.08 0.07 0.05 0.07 0.07

0.06 0.04 0.03 0.05 0.07

0.50 0.03 y0.16 0.04 0.01

0.44 0.08 y0.04 0.04 y0.01

0.32 0.07 y0.03 0.07 0.04

Notes: See Appendix.

ranges from y 0.27% ŽNIGEM, Germany. to 0.02% ŽMULTIMOD, Germany.. The NIGEM results indicate that by year 15 Germany is most adversely affected in terms of output by the oil price shock Žy0.56%., while Japan is least affected Žy0.09%.. The MSG2 model sees the US as being most affected by the shock Žy0.28%. and the UK as being least affected Žy0.09%.. The ranking of responses in MULTIMOD is different yet again, with the Japan most affected by year 15 Žy0.08%. and the US least affected Žy0.02%.. Factor substitution occurring within the production process can soften the impact of the oil price shock on domestic income. As NIGEM has no explicit production function, the absence of factor substitution helps to explain the slightly larger negative effects on GDP within the model. A further negative influence on aggregate demand is the increase in the general price level as the oil price increase

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 13 Sustained increase in world oil price by 25%. Standardised fiscal and monetary rules MSG2

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15





y0.04 y0.21 y0.25 y0.26 y0.28

y0.07 y0.10 y0.11 y0.14 y0.16

y0.05 y0.10 y0.12 y0.19 y0.20

y0.09 y0.05 y0.03 y0.10 y0.09

y0.17 y0.31 y0.39 y0.46 y0.50

y0.17 y0.22 y0.23 y0.22 y0.21

y0.11 y0.14 y0.15 y0.14 y0.11

y0.14 y0.13 y0.12 y0.11 y0.07

y0.09 y0.14 y0.16 y0.18 y0.19

y0.08 y0.10 y0.12 y0.13 y0.13

y0.06 y0.10 y0.12 y0.14 y0.13

y0.07 y0.10 y0.11 y0.12 y0.11

0.02 0.28 0.40 0.45 0.46

0.07 0.15 0.26 0.33 0.35

0.00 0.13 0.26 0.40 0.41

y0.03 0.05 0.21 0.42 0.40

0.25 0.12 0.06 0.01 0.01

0.29 0.06 0.05 0.00 0.00

0.11 0.06 0.06 0.01 0.00

0.05 0.05 0.08 0.01 y0.01

0.02 0.23 0.36 0.41 0.40

0.03 0.09 0.26 0.35 0.36

y0.10 0.03 0.20 0.33 0.33

y0.10 0.02 0.20 0.33 0.33

0.31 0.37 0.40 0.40 0.40

0.23 0.29 0.34 0.36 0.36

0.17 0.25 0.30 0.33 0.34

0.17 0.25 0.30 0.33 0.34

0.22 0.05 y0.15 y0.47 y0.68

0.03 0.15 0.40 0.80 1.04

y0.18 0.14 0.53 1.12 1.49

y0.19 0.14 0.54 1.14 1.50


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 13 Ž Continued. MSG2 US REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15




0.31 0.23 0.04 y0.30 y0.49

0.19 0.02 0.13 0.33 0.45

y0.18 y0.19 y0.11 0.04 0.12

y0.26 y0.31 y0.15 0.14 0.20

0.00 y0.01 0.00 y0.01 y0.03

y0.03 0.01 0.05 0.05 0.04

0.00 0.04 0.08 0.09 0.09

0.03 0.11 0.17 0.21 0.23

0.14 0.18 0.19 0.19 0.21

0.09 0.11 0.11 0.11 0.11

0.08 0.10 0.11 0.12 0.12

0.06 0.07 0.07 0.07 0.07

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

y0.08 y0.02 0.05 0.14 0.12

Notes: See Appendix.

is passed on. Asset holdings of the public fall in real terms, lowering wealth and hence expenditure of domestic consumers. The magnitude of the negative output effect is also influenced by trade relationships with oil-exporting countries. The extra revenue earned by OPEC countries within the models leads to an increase in their imports, reflecting respending. These extra imports are extra exports for many of the industrial countries and help counteract the negative effects on GDP. The size of this positive effect depends essentially on the trade weights between countries, which explains some of the differences in GDP responses across countries within each model. The increase in the world price of oil is initially transmitted to the domestic economy by higher import prices. This effect is then passed on to domestic prices, although the speed and extent of adjustment differs between countries within each model and between models. The impact effect on prices ranges between y 0.19% below base ŽNIGEM, Japan. and 0.44% above base ŽMULTIMOD, US.. The response of prices in NIGEM is greatest in the UK, with the GDP deflator up 1.61% by year 15. The response of prices in Japan is very different to that of the other three countries, as they lie below base throughout the simulation period. In

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Table 14 Sustained increase in world oil price by 25%. Standardised fiscal and monetary rules Žexcept for Japan. NIGEM

GDP 1 3 5 10 15 CON 1 3 5 10 15 INV 1 3 5 10 15 PGNP 1 3 5 10 15 INF 1 3 5 10 15 RS 1 3 5 10 15 RL 1 3 5 10 15 ER 1 3 5 10 15





y0.20 y0.34 y0.41 y0.43 y0.32

y0.11 y0.02 y0.04 y0.10 y0.09

y0.27 y0.39 y0.36 y0.49 y0.56

y0.08 y0.03 y0.02 y0.08 y0.14

y0.20 y0.34 y0.42 y0.45 y0.33

y0.07 y0.13 y0.15 y0.20 y0.25

y0.14 y0.34 y0.48 y0.69 y0.83

y0.04 y0.19 y0.24 y0.20 y0.16

y0.05 y0.14 y0.17 y0.18 y0.14

y0.02 y0.14 y0.25 y0.36 y0.41

y0.11 y0.26 y0.31 y0.43 y0.53

y0.02 y0.10 y0.15 y0.21 y0.29

0.21 0.50 0.60 0.57 0.53

y0.19 y0.14 y0.11 y0.12 y0.17

0.06 0.30 0.26 0.28 0.26

0.27 0.60 0.94 1.50 1.61

0.21 0.08 0.04 y0.02 0.00

y0.20 y0.09 0.05 y0.01 y0.01

0.06 y0.02 y0.01 0.00 0.00

0.28 0.20 0.17 0.07 0.00

0.06 0.08 0.08 0.05 0.05

0.01 0.08 0.09 0.09 0.08

0.04 0.08 0.11 0.12 0.13

0.04 0.07 0.10 0.13 0.14

0.07 0.06 0.06 0.05 0.06

0.08 0.09 0.09 0.09 0.09

0.10 0.11 0.12 0.14 0.19

0.10 0.11 0.12 0.16 0.21

y1.02 y1.03 y0.99 y0.85 y0.76

0.45 0.46 0.41 0.22 0.06

1.29 1.29 1.23 0.97 0.68

1.30 1.31 1.26 0.98 0.63


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Table 14 Ž Continued. NIGEM

REER 1 3 5 10 15 CArGDP 1 3 5 10 15 GDEFrGDP 1 3 5 10 15 M 1 3 5 10 15





y0.71 y0.99 y0.98 y0.98 y0.96

y0.22 y0.38 y0.48 y0.65 y0.78

0.42 0.33 0.12 y0.25 y0.39

0.15 0.21 0.35 0.51 0.44

y0.14 y0.02 0.03 0.10 0.06

y0.15 y0.02 0.02 0.02 0.02

y0.29 y0.13 y0.03 y0.01 y0.04

0.08 0.32 0.43 0.41 0.38

0.12 0.09 0.02 y0.03 y0.01

y0.01 0.05 0.03 0.08 0.07

0.06 0.10 0.02 y0.01 y0.02

0.05 0.05 0.01 0.02 0.00

0.17 0.18 0.20 0.21 0.16

0.14 0.08 0.01 0.01 y0.02

0.14 0.09 0.05 0.04 0.02

0.09 0.39 0.69 1.15 1.25

Notes: See Appendix.

MSG2 the response of prices is greatest in the US Ž0.56%, year 50. and smallest in the UK Ž0.1%, year 50., which can be contrasted with NIGEM, where the UK experiences the largest increase in prices. In MULTIMOD the price response is again greatest in the US Ž0.44%, year 1. and smallest in Japan Ž0.13%, year 1.. Interest rates throughout the simulation target the money base and adjust in response to money supply deviations. The changes in prices and GDP which follow the oil price shock affect the money supply and hence interest rates. The results show that the combined initial effects on prices and GDP provide pressure for the money supply to increase, and so the interest rate rises. The general magnitude of these interest rate effects is largest in MSG2 and smallest in NIGEM, although the positive effects experienced in MULTIMOD are much more short lived than those in the other models. The interest rate changes have implications both for the interest sensitive components of aggregate demand and for exchange rates. Interest rate differentials cause exchange rate movements, which feed through into prices and on to GDP, and thus contribute to determining the GDP response. In NIGEM, despite the ERM assumption, the reported results do not show identical interest rate increases in Germany and the UK, due to variation in

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


implementing the alternative fiscal solvency rules. The interest rate increases further throughout the simulation period, whereas the rise in both Japan and the US tends to remain constant. The response in Germany is consistent with the high rate of increase in prices experienced during the initial stages of the simulation. The relative increase in German interest rates leads to an initial currency appreciation which helps to stabilize German prices. Output in Germany falls by more than in any other country due to the relatively large reduction in the interest-sensitive components of aggregate demand. By the end of the simulation period consumption and investment fall by around 0.85% and 0.55% of base GDP, respectively, with the consumption response being very different to that in the other countries. The interest rate and price increases in the US are not as large as those experienced in the other countries, yet the negative effect on output is larger than that in both Japan and the UK. In the US, the relatively large GDP response is due to trade sector effects rather than the interest-sensitive components of demand. The real effective exchange rate depreciation which results from the interest rate and price changes means that the price of all imports is increasing, not just the price of oil, and these additional price effects amplify the negative GDP response. In MSG2, interest rates rise above base in all four countries, the ranking of the interest rate changes being consistent with both the GDP and price responses, and the Monetary Union assumption leading to a response in the UK identical to that in Germany. Given that the ranking of the interest rate responses is the reverse of that in NIGEM, it is not surprising that the reaction of the exchange rate is also different. In MSG2, US interest rates increase relative to those in the other countries leading to an initial appreciation of the dollar, followed by a depreciation over the simulation period. Although this reaction is the opposite to that in NIGEM, the US dollar depreciates relative to base by the end of the simulation period in both models. Output in the US falls progressively over time and can be explained by both the higher interest rate and the depreciating currency, which implies an increase in the price of all imports. Given the large interest rate response in the US, the negative effects on consumption and investment are much larger than elsewhere and, as in NIGEM, the former outweigh the latter. The increased negative effect on US GDP is primarily due to these private expenditure responses. German and UK interest rate changes result in an initial depreciation of both the deutschmark and the pound, followed by an appreciation over the period, which results in a fall in the price of imports, which in turn puts downward pressure on domestic prices. The negative pressure on prices helps GDP recover slightly towards the end of the simulation period. German GDP returns towards base much more quickly than UK GDP as the negative consumption effects diminish at a much faster rate. The interest rate and price responses in Japan yield a real exchange rate appreciation which increases over time and is greater than that in both the UK and Germany, leading to a fall in competitiveness. The lower import prices which result are transmitted through into domestic prices. By year 15, prices in Japan have increased by an amount which is less than the price increases in the other three countries. Thereafter prices stabilize, while those in the UK and Germany fall back towards base. This is consistent with the result that by the end


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

of the simulation period the negative effect on output is greater than that in both the UK and Germany and is also much slower in returning back to base. Again, the flexibility of the Japanese labour market is apparent as unemployment is lower for 1 year only, returning to base in year 2. The sluggishness of the German labour market is also illustrated by the much slower elimination of the increase in unemployment that follows the shock. As noted above, the interest rate responses in MULTIMOD follow a very different path, with the largest effects occurring during the initial stages of the simulation. Nevertheless the ranking of the interest rate effects is similar to that in MSG2, with the US experiencing the largest increase and Germany the smallest. The increase in US interest rates leads to a dollar appreciation, as in MSG2, although by year 15 the dollar has still appreciated relative to base, whereas in MSG2 there has been a real depreciation. The appreciating dollar in MULTIMOD helps to lower US domestic prices through lower import prices. As prices fall back towards base, the negative effect of the oil price increase on US GDP also diminishes. Initial interest rate changes in Japan and Germany are not as large as the increase in the US rate and therefore result in a depreciation of both their nominal exchange rates. GDP responses in MULTIMOD are not as diverse as those in the other two models. Any short-run differences that do exist across countries are mainly due to different trade balance responses. 3.5. Sensiti¨ity analysis: interest elasticity of money demand The monetary policy rule employed in these simulations sets the short-term rate of interest to achieve the money stock target in the medium term. The transmission of shocks, both nominal and real, through financial markets depends critically on the relationship between nominal demand and the short-term interest rate, which is determined in the first instance by the money demand equation. While the three models all adopt the generic form represented in Eq. Ž6., their different approaches to quantification again result in important differences in specification. The money demand equations in the MSG2 model are static, with an interest rate semi-elasticity of y 0.6. Treated as a long-run value this is more than two standard deviations smaller than both the common long-run value in MULTIMOD’s dynamic equations Žy2.57. and all of the G7 country-specific values in NIGEM. Treated as a short-run or impact elasticity the MSG2 value is within one standard deviation of the corresponding MULTIMOD and NIGEM values for all countries, when measured on an annual basis, but in a static equation the appropriate comparison is with the long-run elasticities of the other equations. To judge the impact of this difference in model specification the simulations of changes in fiscal and monetary policy in the US are re-run with an alternative version of the MSG2 model of the US. The alternative US money demand equation is log Mt s a 1 log Mty1 q Ž 1 y a 1 .w log Pt y log Yt y a 2 RSt x

Ž 6X .

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


where a 1 s 0.8, a 2 s 2.0. The equation has a simple partial adjustment form and an increased long-run interest rate coefficient, corresponding to the value of this coefficient in NIGEM’s US money demand equation. The monetary policy rule Ž7. above is changed conformably. The results of the permanent US fiscal shock in MSG2 under the alternative money demand function are shown, for several variables, in the final column ŽMSG2U . of Table 1. It is clear from a comparison of interest and inflation rate responses that the greater interest rate elasticity in the alternative equation makes the initial expansionary effect of fiscal policy more pronounced. The domestic US impact multiplier is now 0.64, compared with 0.33 under the original specification. Increased sensitivity of the interest rate to the money target allows monetary policy to be more accommodating. Inflation is also appreciably higher in the early years of the simulation but the real short-term interest rate is lower for 10 years. These responses are now more in line with those of the other models. For the monetary shock, the corresponding results are shown in the final column of Table 8. As a result of the increased interest elasticity of the demand for money, the depreciation of the exchange rate following the monetary shock is muted, as is the associated expansion in GDP. However, the initial positive output effect remains in excess of that for the other models, due to the immediate expansion of net trade flows. If the alternative money demand equation is restricted to be static, but the higher interest rate semi-elasticity is retained, then the simulation results are unchanged to the second decimal place from those of the initial alternative model. Thus it is the size of the interest rate semi-elasticity rather than the lack of dynamics that is the crucial factor. We finally note, however, that the assumed monetary policy regime dramatises the impact of this model specification question, and under other possibilities, such as nominal income targeting or combinations of price and real income targeting, its impact is reduced. The reason is that money itself plays a minor role in the MSG2 model, as in the other two models. It appears as a component of the financial and physical wealth of households and thereby has a direct impact on consumption. Differences in the form of the money demand equation, however, have only a negligible impact on the results of a fiscal policy simulation such as those in Table 1 under nominal income targeting, since changes in personal sector holdings of money are small and so have little impact on consumption. This does not mean that monetary policy is unimportant, rather that the impact of interest rate changes is crucial to model properties in all cases.

4. Conclusion A first conclusion, relative to the existing literature on global economic model comparisons, is that hands-on access to the models allows a fuller appreciation of the source and nature of differences between them to be gained. This inevitably requires much detailed scrutiny of model structures and properties, nevertheless


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

some broad similarities and some important differences emerge from the detail of the simulation results. Going beyond the particular explanations provided in the foregoing account, a further conclusion is that the general approach to quantification is a basic source of differences between the present models. Many similarities in their theoretical foundations exist, but the variation in the relative weight given to theory and empirical evidence leads to greater dissimilarity in the outcomes. The present models could be commended for their ‘modernity’ in the same spirit as that in which Sims has recently commended a further model ŽTaylor, 1993. not included in our study. It ‘uses consumption and investment functions loosely based on dynamic optimizing theories and incorporating expectations of future income explicitly. It has wage adjustments that are sluggish and depend on expectations of the future. It has international linkages that use expectational interest rate parity conditions. In these respects it is an advance beyond the state of the art in 1980’ ŽSims, 1996, p. 116.. Nevertheless Sims goes on to criticise the statistical methods employed and it is at the same point that our present models diverge. MSG2 takes the dynamic optimizing theories more seriously and the model is calibrated in the CGE tradition; where econometric estimates are drawn upon this is at second hand, and these estimates have seldom been produced for the purposes of large-scale macroeconomic modelling. MULTIMOD and NIGEM place greater weight on the time series evidence, remaining in what Sims refers to as ‘the original simultaneous equation modelling tradition’. Relatively short series of annual economic aggregates are often relatively uninformative about parameters that the theory says are important, however, and a further divergence occurs here. MULTIMOD improves the precision of estimation by pooling evidence across countries, at a potential cost of eliminating cross-country behavioural differences. NIGEM seeks more information in quarterly data and allows countries to differ; where imprecision remains, in the sense of statistical insignificance, coefficients are set to zero, which may represent too great a weight on hypothesis testing rather than estimation, within an approach heavily weighted to the empirical evidence. Many of the differences we have documented originate in these differences of approach, hence our study also provides a case study within the kind of methodological debate to which Sims’ article is principally oriented.

Acknowledgements The research reported in this article was supported by the Economic and Social Research Council Žref L120251018. as part of its Global Economic Institutions Research Programme. The authors are grateful for the cooperation of the model proprietors, Ray Barrell ŽNIESR., Warwick McKibbin ŽANU. and Steve Symansky ŽIMF., and for comments received following a preliminary presentation of results at the ESRC Macroeconomic Modelling Seminar held at the University of Warwick on 4]6 July 1995. An anonymous referee also provided helpful comments.

P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48


Appendix 1 Notes for Tables 1]14

s gross domestic product measured in real terms, percentage deviation from base; CON s total private consumption, deviation from base, expressed as percentage of baseline GDP; INV s total private investment, deviation from base, expressed as percentage of baseline GDP; PGDP s gross domestic product deflator, percentage deviation from base; INF s inflation rate in gross domestic product deflator, percentage point deviation from base; RS s short term interest rate, percentage point deviation from base; RL s long term interest rate on government bonds, percentage point deviation from base; ER s nominal US Dollar exchange rate, percentage deviation from base; REER s real effective exchange rate, percentage deviation from base; CArGDP s current accountrgross domestic product ratio, percentage point deviation from base; GDEFrGDP s government budget deficitrgross domestic product ratio, percentage point deviation from base; M s money stock, percentage deviation from base; and MSG2U s results obtained under alternative MSG2 specification, discussed in Sec. 3.5. GDP

References Barrell, R.J., Whitley, J.D. ŽEds.., 1992. Macroeconomic Policy Coordination in Europe: the ERM and Monetary Union. Sage Publications, London. Bryant, R.C., Henderson, D.W., Holtham, G., Hooper, P., Symansky, S. ŽEds.., 1988. Empirical Macroeconomics for Interdependent Economies. Brookings Institution, Washington, DC. Bryant, R.C., Hooper, P., Mann, C.L. ŽEds.., 1993. Evaluating Policy Regimes: New Research in Empirical Macroeconomics. Brookings Institution, Washington, DC. Church, K.B., Mitchell, P.R., Sault, J.E., Wallis, K.F., 1996. The sensitivity of global model simulations to fiscal closure rules. ESRC Macroeconomic Modelling Bureau Discussion Paper No.45, University of Warwick. Fischer, S., 1977. Long-term contracts, rational expectations, and the optimal money supply rule. J. Political Econ. 85, 191]205. International Monetary Fund, 1991. Changes to MULTIMOD since the July 1990 Occasional Paper No. 71. Internal working paper, Research Department, IMF. Layard, R., Nickell, S., Jackman, R., 1991. Unemployment, Macroeconomic Performance and the Labour Market. Oxford University Press, Oxford. Masson, P.R., Symansky, S., Haas, R., Dooley, M., 1988. MULTIMOD: A multi-region econometric model. Staff Studies for the World Economic Outlook, July 1988. International Monetary Fund, Washington, DC, pp. 50]104.


P.R. Mitchell et al. r Economic Modelling 15 (1998) 1]48

Masson, P.R., Symansky, S., Meredith, G., 1990. MULTIMOD Mark II: A revised and extended model. Occasional Paper No.71, International Monetary Fund, Washington, DC. McKibbin, W.J., Sachs, J.D., 1991. Global Linkages: Macroeconomic Interdependence and Cooperation in the World Economy. Brookings Institution, Washington, DC. Sims, C.A., 1993. Comments and reflections. In: Bryant, R.C., Hooper, P., Mann, C.L. ŽEds.., Evaluating Policy Regimes: New Research in Empirical Macroeconomics. Brookings Institution, Washington, DC, pp. 430]443. Sims, C.A., 1996. Macroeconomics and methodology. J. Econ. Perspect. 10, 105]120. Taylor, J.B., 1979. Estimation and control of a macroeconomic model with rational expectations. Econometrica 47, 1267]1286. Taylor, J.B., 1993. Macroeconomic Policy in a World Economy. Norton, New York. Turner, D.S., Richardson, P., Rauffet, S., 1993. The role of real and nominal rigidities in macroeconomic adjustment: a comparative study of the G3 economies. OECD Econ. Stud. 21, 90]137. Wallis, K.F., 1993. Comparing macroeconometric models: a review article. Economica 60, 225]237.