The macroeconomic effects of higher oil prices.

Hunt, Benjamin ; Isard, Peter ; Laxton, Douglas 等

Douglas Laxton (*)

The paper uses MULTIMOD to analyse the macroeconomic effects of oil price shocks, distinguishing between temporary, more persistent, and permanent shocks. It provides perspectives on several findings in the literature and the key role of monetary policy in influencing macroeconomic outcomes. Specific attention is paid to the channels through which oil price increases can pass through into core inflation, the implications of delayed policy responses, and the relative merits of leaning in different directions when the correct policy response is uncertain.

Introduction

After reaching a 25-year low in February 1999, oil prices rose sharply over the next two years. Chart 1 plots an historical series for the real price of oil and is useful for putting this price hike into an historical context. Corresponding to the spikes in Chart 1, the real price of oil quadrupled during the first oil price shock of the 1970s, tripled during the shock at the end of the 1970s, doubled during the second half of 1990, and tripled during 1999-2000. Given the macroeconomic developments that followed the oil shocks of the 1970s, the substantial rise in oil prices during 1999 and 2000 generated concerns about the prospects for world growth and inflation and integrally-related questions about the appropriate way for monetary policy to respond.

In the event, these concerns were overtaken by the pronounced weakening of the global economy during 2001 and the associated sharp decline in oil prices. Nevertheless, it is important to pursue a deeper understanding of the issues. This paper uses the IMF's multicountry model, MULTIMOD, to analyse the macroeconomic effects of oil price shocks, with particular focus on the implications for economic activity and inflation in the industrial countries. The analysis provides perspectives on several findings in the literature, and on the role of monetary policy in influencing outcomes. While the focus is on the effects of increases in oil prices, the framework can be used symmetrically to analyse the effects of oil price declines.

The macroeconomic turbulence that followed the two major oil price shocks of the 1970s stimulated a large body of research that has attempted to estimate the effects of oil price shocks on inflation and economic activity. (1) This research has found clear negative correlations between oil prices and aggregate measures of economic activity, as well as significant correlations between oil prices and microeconomic data on output, employment, and real wages. In addition, there is strong evidence of asymmetry in the relationship between oil price changes and subsequent changes in economic activity.

Empirical research has generated evolving impressions about the magnitude of oil-price effects on aggregate economic activity and about the extent to which activity responds symmetrically to oil-price increases and oil-price declines. The empirical evidence presented in Hamilton (1983), based on linear VAR models, suggested that exogenous shocks to oil prices had significant effects on real economic activity in the United States. Subsequently, the fact that the large decline in oil prices in the mid-1980s did not result in an output boom seemed to suggest that the relationship had changed. Mork (1989) extended the work of Hamilton, allowing oil price shocks to have asymmetric effects and inferring that oil price increases reduced real output while oil price declines had no effect. Several years later, using data up to 1994, Hooker (1996) concluded that the relationship uncovered by Hamilton had broken down and that allowing for asymmetric output response to price increases and price decreases did not alter th at result. More recently, Hamilton (2000) has provided clear evidence of nonlinearity - "oil price increases are much more important than oil price decreases, and increases have significantly less predictive content if they simply correct earlier decreases." (2)

Many economists believe that monetary policy is responsible for generating the observed negative correlation between oil prices and economic activity, and may have played a role in contributing to the apparent instability of the correlation over time. Bernanke and others (1997) provided analysis suggesting that monetary policy has been the primary reason that oil price increases have had negative output effects in the United States. (3) Needless to say, the rationale for monetary policy responses to exogenous increases in oil prices is to contain the effects on inflation. In that connection, however, recent work by Hooker (1999) suggests that, in the United States since 1981, oil price shocks have only affected headline inflation, with no impact on core inflation. Do these results suggest that monetary policy no longer needs to respond to oil price innovations? According to the nonlinear relationship between oil-price movements and output as outlined in Hamilton (2000), the sharp rise in oil prices over the l ast few years could lead to a decline in real output growth. However, if Bernanke and others (1997) and Hooker (1999) are correct, an output decline might appear to be avoidable. In particular, if core inflation does not respond to oil price increases, then there might be no need for monetary policy to tighten, in which case the effects on real economic activity could be minimal.

MULTIMOD simulations can help shed light on these issues and provide more general perspectives on both the key channels through which exogenous oil price innovations affect the macroeconomy and the associated implications for monetary policy. The main points that our simulations are designed to illustrate are the following.

First, even if oil price shocks pass through into core inflation, the response to a temporary oil price increase can look very similar to the response that would be observed in an economy with no pass-through of oil price shocks into core inflation. This result suggests that the empirical evidence from the 1980s and 1990s, as analysed by Hooker (1999), needs to be interpreted with caution. One possible reason why the data seem to suggest that oil price shocks no longer have an impact on core inflation may be the fact that the positive innovations to oil prices during the 1980s and most of the 1990s were very short lived. A second possibility is that monetary policy may have reacted differently to oil shocks during the 1980s and 1990s than it did during the 1970s. In that connection, Bernanke and others (1997) concluded that the declines in US output following the 1979 and 1990 oil-price shocks were largely a result of monetary policy, whereas the recession in 1974-5 was primarily due to factors other than mon etary policy. To the extent that this was the case, however, it would not be valid to interpret Hooker's (1999) findings as providing a rationale for monetary policy behaviour to change again by no longer responding to oil price increases. As Lucas (1976) has emphasised, there are dangers in assuming that estimated reduced-form relationships are invariant to changes in the behaviour of policy.

A second point that we address is the extent to which the observed nonlinearity in the relationship between oil prices and macroeconomic activity might be attributable to asymmetries at the microeconomic level. In particular, we show that if most of the output effects arising from oil price shocks are associated with the monetary policy response, then asymmetric pressures on core inflation, arising perhaps from asymmetric responses of microeconomic agents to the impact effects of oil price changes on their real incomes, could help explain the asymmetric response of economic activity to oil price changes.

We also use MULTIMOD to illustrate that, even when an oil price increase turns out to be persistent and core inflation responds, a slow reaction by policymakers will not necessarily magnify the macroeconomic implications. This raises the possibility that policymakers may have the luxury of waiting to respond until they see clear evidence that core inflation is increasing. We emphasise, however, that the scope for delay depends critically on whether the slow policy response leads private agents to doubt the inflation-fighting resolve of central banks. This is illustrated with a simulation showing that the economic dislocation arising from the shock can be magnified if the slow response and the resulting deterioration in inflation performance lead to (temporary) erosion of policymakers' credibility.

Finally, to provide some perspective on how policymakers should respond in the face of uncertainty about wage/price behaviour, MULTIMOD simulations are used to compare the costs of two types of policy errors in responding to an oil price hike. The first error results from policymakers initially assuming that the oil price increase will have no core inflation effect when core inflation in fact responds positively and asymmetrically to changes in oil prices, and when monetary policy credibility can be eroded. The second error is the result of policymakers initially believing that agents will respond in the most inflationary manner when in fact they respond in the most benign manner. Comparing the estimated costs of these two errors suggests that, other things being equal, policymakers might want to lean in the direction of high-side assumptions about the extent to which persistent oil price increases lead to core inflation pressures.

The remainder of the paper is structured as follows. The next section presents some stylised facts about the behaviour of oil prices. It is followed by a brief outline of the structure of MULTIMOD and the channels through which oil price movements can influence the macro-economy. Simulations of the impacts of oil prices under different behavioural assumptions are presented in the fourth section, along with comparisons of the costs of making the two alternative monetary policy errors described above. Finally, some conclusions are presented in the last section.

Oil prices

Two measures of oil price changes since the early 1970s are presented in Chart 1. The first is the quarterly per cent change in the US dollar price of oil and the second is a net oil price increase series proposed in Hamilton (1996 and 2000). The latter series measures the amount by which oil prices in a given quarter exceed their peak value over the previous twelve months; if they do not exceed the previous peak, the measure is set to zero. Hamilton (2000) presents evidence suggesting that his proposed measure performs significantly better in predicting the impact of oil price changes on real economic activity. A number of points are worth noting about the behaviour of oil prices suggested by these measures.

First, the quarterly per cent change in oil prices shows that, between their sharp decline in the mid- 1980s and 1998, real oil prices have exhibited several upward spikes. These notable increases, however, have been very short-lived. This contrasts considerably with the behaviour of oil prices in the 1970s, when two major increases turned out to be very persistent. Second, the measure proposed in Hamilton (1996 and 2000) shows less variability between 1982 and 1998 as it filters out episodes of short-lived increases. However, it suggests that the episode of oil price increases that began in early 1999 has the potential to have significant macroeconomic effects.

In addition to the two points above, analysis presented in Cashin and others (1999) indicates that the historical behaviour of oil prices does not allow one to predict how future oil price cycles will evolve. The severity of price movements provides no information about their likely duration and the time spent in a current boom or slump provides no information about the likely future duration of that boom or slump. Taken together these characteristics of oil prices suggest that, even though it is difficult to predict how long high oil prices might persist, the increase that occurred between early 1999 and the end of 2000 could have a significant impact on economic performance. In the simulation analysis that follows, MULTIMOD is used to illustrate the potential macroeconomic implications of oil price shocks similar in magnitude to the recent experience. These simulations consider of price shocks of various durations as well as alternative assumptions about the degree of pass-through into core inflation, the p olicy response, and the possible effects on private agents' perceptions about monetary policy objectives.

The transmission of oil price shocks in MULTIMOD

An oil price increase can influence macroeconomic behaviour through several channels. Five of these seem particularly relevant in the first few years following the shock. First, the transfer of income from oil-importing countries to oil-exporting countries is expected to reduce global demand as demand in the oil-importing countries is likely to decline more than it will rise in the oil-exporting countries. This reflects an assumption that the propensity to spend in the oil-exporting countries is likely to be significantly smaller in the short run than in the oil-consuming countries. Second, the increase in the cost of inputs to production can reduce the amount of non-oil (potential) output that can be profitably supplied in the short run, given the existing capital stock and assuming that wages are relatively inflexible in the short run. Third, workers and producers may resist declines in their real wages and profit margins, putting upward pressure on unit labour costs and the prices of finished goods and ser vices. Fourth, the impact of higher energy prices on headline price indexes (e.g., consumer price levels) and the potential for pass-through into core inflation may induce central banks to tighten monetary policy. And fifth, to the extent that policy reactions seem inconsistent with announced policy objectives, the credibility of the monetary authorities may be eroded, with consequences for inflation expectations and the inflation process. (4)

MULTIMOD is a multi-regional macroeconometric model developed by the IMF staff for the primary purpose of analysing alternative scenarios for the World Economic Outlook (WEO). As such, it is based on annual data and takes the WEO forecast as an 'exogenous' baseline. Its construction has gone through several stages. To conserve on space the simulation results presented in this paper focus only on the United States and the Euro Area. These results have been based on an extended Mark III version of MULTIMOD. (5) Modern structural models like MULTIMOD have been designed to minimise first-order Lucas-critique problems and to provide insights into the key role of the monetary policy response in influencing the macroeconomic effects of various exogenous shocks. (6)

MULTIMOD-based analysis of oil-price shocks hinges critically on the nature of wage/price behaviour and the monetary policy reaction function. The former is described in detail at the end of this section. In the extended version of MULTIMOD monetary policy is characterised by an inflation-forecast-based (IFB) rule. (7) As elaborated below, the monetary authorities are assumed to set the short-term interest rate in response to new information about their forecast of inflation and the magnitude of the output gap.

In characterising the supply side of the economy, MULTIMOD assumes that production technology uses capital and labour inputs with no explicit role for inputs of primary or intermediate products. (8) Firms choose the profit maximising level for the capital stock based on production technology, input costs, and output prices. In MULTIMOD, the aggregate absorption price deflator approximates input costs and the GDP deflator represents the aggregate output price. Consequently, permanent oil price shocks drive a wedge between input and output prices that reflects the country's dependence on net imports of oil; and firms then adjust their desired capital stock accordingly. The new capital stock is achieved through adjustment in investment flows. In the long run, the level of potential output will reflect the new level of the capital stock. In reality, however, capital stock adjustment may occur much more rapidly than MULTIMOD suggests, as firms have the ability to retire quickly capital that becomes relatively inef ficient to operate following an increase in oil prices perceived to be permanent. Costly reallocation of labour and capital across sectors and the discouraging effect of increased uncertainty on irreversible investment (behavioural features that MULTIMOD may not adequately capture) have also been cited in the literature as potentially important supply-side effects of oil price increases. These considerations suggest that MULTIMOD may underestimate the short-run effects of positive oil price shocks on output. Elsewhere we have judgementally adjusted potential output to allow for such effects, (9) but for present purposes this is not a matter of concern. (10)

The effects of oil price shocks on CPI and core inflation in MULTIMOD

This section describes the main channels through which oil price shocks can have direct inflationary effects in the Mark IIIA version of MULTIMOD. The discussion focuses on those equations of the model that play a key role in transmitting the effects of oil-price increases into the inflation process in the major industrial countries/blocks.

MULTIMOD, like most macroeconomic policy models, relies on a reduced-form Phillips curve to characterise the behaviour of inflation in the industrial countries. (11) The model of inflation and inflation expectations distinguishes between CPI inflation and core inflation, where core inflation is defined as the rate of change in the GDP deflator excluding oil and is taken to be the measure on which monetary policy decisions are based. Although MULTIMOD does not include explicit wage rates, the dynamics of inflation and inflation expectations are characterised in a manner that implicitly recognises important features of wage-setting behaviour (in particular, contracting lags and wage-push elements), and these equations are sometimes referred to as the wage/price nexus.

The key equations in MULTIMOD'S reduced-form wage/price structure are

[[pi].sup.CPI.sub.t] = [[delta].sub.1][[pi].sup.M.sub.t] + [[delta].sub.2][[pi].sup.C.sub.t] + [[delta].sub.3][[pi].sup.POIL.sub.t]

+[1 - [[delta].sub.1] - [[delta].sub.2] - [[delta].sub.3]][[pi].sup.CPI.sub.t-1] (1)

[[pi].sup.C.sub.t] = [psi][[pi].sup.e.sub.t+1] + [1 - [psi]][[pi].sup.C.sub.t-1]

+[gamma][([u.sup.*.sub.t] - [u.sub.t])/([u.sub.t] - [[phi].sub.t])]

+[alpha][[[pi].sup.CPI.sub.t-1] - [[pi].sup.C.sub.t-1]] (2)

[[pi].sup.e.sub.t+1] = [OMEGA][[lambda][[pi].sup.CPI.sub.t+1] + (1 - [lambda])[[pi].sup.c.sub.t+1]]

+(1 - [OMEGA])[[gamma][[pi].sup.CPI.sub.t-1] + (1 - [lambda])[[pi].sup.c.sub.t-1]] (3)

Here [[pi].sup.CPI] is CPI inflation; [[pi].sup.M] is the rate of inflation of the domestic-currency price of manufactured imports; [[pi].sup.POIL] is the rate of inflation of the domestic-currency price of oil; [[pi].sup.C] is core inflation (non-oil GDP deflator); [[pi].sup.e] is a measure of expected inflation; [u.sup.*] is the non-accelerating-inflation rate of unemployment (the NAIRU); u is the unemployment rate; [empty set symbol] is the minimum absolute lower bound for the unemployment rate; and [psi], [alpha], [gamma], [OMEGA], [lambda], [[delta].sub.1], [[delta].sub.2] and [[delta].sub.3] are parameters.

Table 1 reports the values of the parameters in the model that are critical for understanding the more direct channels of pass-through of oil prices into both CPI inflation and core inflation. (12) In particular, it reports estimates of the parameter values 10[[delta].sub.3], [[delta].sub.2], [[delta].sub.1], [psi], [alpha], and [lambda] for each country/block, as well as average values for these parameters across all of the industrial country blocks.

Direct contemporaneous effects of oil price shocks on the CPI

The direct contemporaneous effect of a change in oil prices on CPI inflation is measured by the parameter [[delta].sub.3] in equation (1). For presentational purposes, the values of this parameter in Table 1 have been multiplied by a factor of 10 so that they can be reported with the same number of digits as the other parameter values. As can been seen in the table, the direct contemporaneous effects of an increase in [[pi].sup.POIL] on [[pi].sup.CPI] are significantly higher than the average value in the United States and the Euro Area; and the effects are significantly smaller than the average parameter estimate in Japan and the United Kingdom.

Based on the estimates of [[delta].sub.3], Table 2 reports the contemporaneous direct effects on annual CPI inflation that would result from a 50 per cent increase in the price of oil. These estimates suggest that a 50 per cent increase in the price of oil would have a direct positive effect on annual CPI inflation of 1.3 percentage points in both the United States and the Euro Area, 0.6 percentage points in both Japan and the United Kingdom; 0.8 percentage points in Canada and 0.7 percentage points in the block of other industrial countries. (13) Do such estimates seem plausible?

One common approach used to assess the plausibility of econometric estimates of the direct effect of oil price shocks on the CPI is to compare them with estimates derived from a more mechanical direct-accounting approach. This direct-accounting approach is usually based simply on the weights of gasoline and other petroleum products in the CPI baskets of these countries. For most countries, the MULTIMOD estimates are slightly larger than these weights, consistent with the fact that in some of these countries, increases in the price of petroleum products may result in increases in prices of other energy sources such as electricity and natural gas.

Direct dynamic effects of oil price shocks on the CPI and core inflation.

The equations described above can be used to study the direct dynamic effects of oil prices on both CPI and core inflation. The structure of MULTIMOD's inflation block allows for oil price movements to flow into core inflation [[[pi].sup.C]] through two possible channels.

The first channel is an expectations-channel with the potential impact given by the parameter [lambda]. Indeed, as can be seen in equation 3, if [lambda] was equal to zero the inflation expectations variable [[[pi].sup.e]] that enters the core inflation equation would depend entirely on expected changes in the non-oil GDP price deflator and there would be no role for the CPI to influence core inflation through expectations. This extreme case might seem completely unrealistic given that many contracts are negotiated in terms of the CPI. The estimates in Table 1 suggest that the average value of [lambda] is around 0.5, with the individual estimates ranging from a high of 0.74 in the block of other industrial countries to a low 0.31 for Japan.

The second channel is measured by the parameter [alpha] and represents the degree of real-wage catch-up in the bargaining process. For example, a value of [alpha] equal to zero would imply that workers do not attempt to resist reductions in their real consumption wage. The average value for [alpha] is 0.26 in MULTIMOD and ranges from a low of 0.09 in Japan to a high of 0.42 for the United Kingdom.

To illustrate the nature of pass-through to core inflation in MULTIMOD, Chart 2 reports some dynamic impulse response functions for [[pi].sup.CPI] and [[pi].sup.C] for the same oil price shock discussed earlier (a permanent 50 per cent increase) under different assumptions about [alpha] and [lambda]. These estimates of impulse response functions are based on three additional assumptions. First, it is assumed that the oil price shock is an innovation to the world price of oil and that changes in the domestic price of crude oil, measured by [[pi].sup.OIL], also reflect any induced changes in the nominal exchange rate that might result from the oil price shock. Second, for the purpose of this experiment, it is assumed that both the unemployment gap [[u.sup.*] - u] and the real exchange rate are fixed. The latter assumption is implemented by adjusting import prices and the domestic price of oil one-for-one with any change in the non-oil price deflator. These assumptions are obviously unrealistic, but they are us eful for illustrating some of the key linkages in the model.

No pass-through into core inflation

The short-dashed lines in Chart 2 report the impulse response functions for [[pi].sup.CPI] and [[pi].sup.C] when [alpha] and [lambda] are set equal to zero and the other parameters are set at their average values for the industrial countries. In this case, because there are no catch-up effects [[alpha] = 0] or any effects of changes in CPI inflation on the expected inflation term [[lambda] = 0] there are no effects on core inflation. Consequently, CPI inflation rises by 0.9 percentage points in the first year and then reverts back to control very quickly. Under this choice of parameter values, market participants implicitly believe that the change in the oil price will require changes in relative prices in the economy without any significant change in core inflation, and that workers will not resist the relative price changes. Obviously, under these optimistic assumptions, it would not be necessary for monetary policymakers to tighten real monetary conditions (and create an excess-supply gap in the labour mar ket) to contain pressures on core inflation.

Base-case pass-through into core inflation in

MULTIMOD

The solid lines in Chart 2 report the impulse response functions for [[pi].sup.CPI] and [[pi].sup.C] when [alpha] and [lambda] are equal to 0.26 and 0.48 respectively -- the average values in MULTIMOD. Under these parameter values, there would be significant long-term effects on both CPI inflation and core inflation if policies were successful in holding both the real exchange rate and the unemployment gap fixed. Indeed, in this case even the direct impact effect on the CPI is 0.1 percentage point greater than the case where [alpha] and [lambda] are equal to zero. This reflects the fact that inflation expectations are determined partly by a model-consistent component that increases in response to the increase in oil prices. As can be seen in the second panel of chart 2, these assumptions result in a gradual increase in core inflation by 0.8 percentage points within three years, and both CPI and core inflation stabilise at rates that are permanently higher by 0.8 percentage points. Under these assumptions abou t pass-through, an attempt by the monetary authorities to offset the deleterious effects on real activity by holding the unemployment gap fixed would result in an ongoing wage-price spiral and a permanently higher rate of inflation. A clear implication is that when there is any significant pass-through into core inflation, monetary policy must at some point tighten real monetary conditions if it wants to avoid a permanent increase in inflation, other things being equal. Chart 2 also includes an intermediate case where inflation expectations are assumed to be determined partly by CPI inflation [[lambda] is still 0.48], but the real-wage catch-up term has been turned off [[alpha] is imposed to be zero]. As can be seen in the chart, the long-run effects on inflation are about 0.4 percentage points, or about one half of the magnitude of the base-case result in which the two channels are functioning.

The impact of this shock on individual countries ill depend on how their parameter values compare with the average. The parameters reported in table 1 indicate ate that the United States and the Euro Area will experience larger than average effects, the United Kingdom and other industrial countries will experience close to a average effects, with Japan and Canada experiencing below average effects. The simulation analysis presented in the remainder of the paper will focus exclusively on the United States and the Euro Area, the two largest economic areas in the world and the cases in which the largest impacts are felt. The results for a broader rang of countries are reported in Hunt, Isard and Laxton (2001).

MULTIMOD simulations

This section presents several sets of MULTIMOD simulations. The first set of simulations describes the responses of real GDP and inflation to oil-price shocks of different duration, based on the estimated parameters of the model. The second compares the model's responses to transitory and more persistent oil-price shocks under two alternative structures of the wage/price nexus. The third set of simulations illustrates how asymmetric responses by microeconomic agents to changes in their real wages might help explain the observed nonlinear relationship between oil prices and macroeconomic activity. The fourth explores the implications of delaying the monetary policy response to a persistent increase in oil prices under both exogenous private-sector perceptions about the objectives of monetary policy and an alternative formulation that makes those perceptions endogenous. The last set of simulations explores the policy implications of uncertainty about the nature of estimated behavioural relationships.

Responses to shocks of different duration

We first consider how real GDP and inflation would be likely to respond to oil-price shocks of three different durations. Under the first oil-price innovation, here referred to as a temporary shock, the price of oil increases by 50 per cent in the first year and returns to baseline in the second year. The second innovation is a more persistent shock, with oil prices increasing to 50 per cent above baseline for the first two years and then declining at a steady rate that brings them back to the baseline level in the sixth year. The third shock involves a permanent 50 per cent increase in oil prices. The analysis implicitly assumes that the behaviour of futures prices allows market participants and policy authorities to identify correctly the types of oil price shocks to which they are responding.

As noted above, the modelling of inflation and inflation expectations in MULTIMOD distinguishes between CPI inflation and core inflation (i.e., the rate of change in the GDP deflator excluding oil) and includes two separate channels through which the impact effects of oil price shocks on CPI inflation can pass into core inflation. These two channels, and the monetary policy reaction functions, have an important influence on the simulation results reported below.

MULTIMOD's base-case monetary policy reaction functions are forward-looking inflation-forecast-based (IFB) rules. Specifically, the nominal short-term interest rate is adjusted -- relative to an equilibrium nominal interest rate -- in proportion to the deviation of observed output from potential output and the deviation of forecast core inflation from an inflation target. (14) The choice of IFB rules rather than conventional Taylor rules -- which look similar to IFB rules in most respects but focus on the deviation from target of current inflation rather than forecast inflation -- reflects a view that central banks are indeed forward looking in their policy deliberations. It also reflects formal analysis indicating that conventional Taylor rules are not as effective in maintaining macroeconomic stability in a world in which behaviour is moderately nonlinear and private agents form their expectations in a (partially) forward-looking manner. (15)

Chart 3 presents the simulated outcomes for real GDP, GPI inflation, and core inflation in the United States and the Euro Area under the three different shocks. (16) The three shocks have similar first-year effects on CPI inflation, which closely reflect the weights of oil in the two countries' CPIs. Inflation subsides more gradually under the permanent shock than under the persistent shock, while the temporary shock leads to below-baseline inflation when oil prices decline in year two and for a period thereafter. The similarity of the responses in the United States and the Euro Area reflects the combination of a significantly higher degree of resistance to real-income declines in the United States and a significantly greater responsiveness of expectations to oil-price changes in the Euro Area. The important points worth noting are that the magnitude of the effect on real activity and the persistence of the effect on inflation depend on the shock's duration.

The magnitudes of the simulated effects on GDP and inflation should be regarded as illustrative. The impact effects on CPI inflation are realistic, but the other estimates -- while reasonably plausible -- are obviously sensitive to MULTIMOD's descriptions of the wage/price nexus and the behaviour of monetary authorities. It may be noted here again, as discussed in the third section, that several considerations suggest that MULTIMOD may somewhat underestimate the short-run supply-side effects of oil-price shocks on output. In other simulations, potential output has been judgementally adjusted to allow for these effects, but that is not done in this paper, since the main interest here is in relative magnitudes and qualitative results. (17)

The strength of the pass-through into core inflation

The simulations presented in this section are intended to provide perspectives on one of the puzzles that has emerged in empirical investigations of the effects of oil-price shocks -- in particular, the finding that oil price shocks during the 1980s and 1990s had little apparent influence on core inflation in the United States. (18) For this purpose we present simulations that combine two sets of assumptions about the wage/price nexus with two sets of assumptions about the time profile (duration) of the oil-price shock. The first version of the wage/price nexus 'turns off' the channels that allow oil price shocks to pass through into core inflation, the second version allows these channels to operate under the estimated parameters of the model, and the two shocks correspond to the temporary shock and the more persistent shock defined earlier.

The simulation results are reported in table 3. The main conclusion that we draw from these simulations is that when the shock is temporary, the responses of output and inflation under the two different wage/price structures are similar; the most significant difference is in the core inflation outcome for the United Sates because it has a relatively large estimated real-wage catch-up effect. This point deserves emphasis when interpreting empirical evidence on the effects of oil-price shocks during the 1980s and 1990s. Many of the shocks to oil prices that occurred during those decades lasted only one or two quarters -- less than the one-year duration built into the MULTIMOD simulations. While it se ms reasonable to assume that downward pressures on real incomes that last for a year or longer would star to have observable effects on the outcomes of the wage-bargaining process, the real-wage catch-up effect of oil-price shocks may not have been important in the ate 1980s and 1990s, when oil-price shocks were ve ry short-lived. (19) In a world with many other shocks occurring as well, it is easy to understand how attempts to estimate reduced-form Phillips curves might have trouble distinguishing between alternative structures of the wage/price nexus. (20)

Asymmetry in real-income catch-up effects

There is clear empirical evidence of a nonlinear relationship between oil prices and aggregate measures of economic activity: oil prices and economic activity are negatively correlated, but oil price increases tend to be followed by larger changes in activity than oil price declines. It is also widely believed, with some support from empirical evidence, that monetary policy has played an important role in generating the observed relationship.

One hypothesis that could provide a consistent explanation would be that workers respond asymmetrically to oil-price increases and oil-price declines, pushing for higher wages to resist the declines in their real consumption power that result from positive oil price shocks, but not resisting the increases in real consumption power that result from oil price declines. This hypothesis, which is analogous to the popular notion that nominal wages are flexible upwards but sticky downwards, would also provide a rationale for policy reacting asymmetrically to oil price shocks, and for consequent asymmetry in the behaviour of macroeconomic activity.

To examine how well the observed relationship between oil prices and macroeconomic activity can be explained by asymmetry in the real-wage catch-up effect, we present the model's simulated responses to both positive and negative oil price shocks under an asymmetric response to changes in households' real consumption wage. In this specification of the wage/price nexus, workers resist only declines in their real consumption wage. (21) To illustrate the effects of this asymmetry in real-income catch-up effects, we focus on permanent 50 per cent changes (increases and decreases) in the price of oil. (22)

Table 4 presents the simulation results. It is readily apparent that the asymmetry in the real-wage catch-up effect generates asymmetry in the responses of output and inflation to the positive and negative oil price shocks. It is also evident that the degree to which output and inflation respond asymmetrically to oil price increases and decreases -- as indicated by the ratio of the effects of the positive shock to the effects of the negative shock -- varies slightly between the United States and the Euro Area. Since the transmission of oil-price effects through the expectations-channel is modelled symmetrically, the different degrees of asymmetry between the two countries can be attributed largely to the differences in the strength of the real-wage catch-up effect. This is apparent in the fact that the asymmetries are more pronounced for the United States than for the Euro Area. It may be noted that asymmetry in the estimated real-wage catch-up effect does not, by itself, support the extreme view that positiv e oil-price shocks have contractionary effects while negative oil-price shocks have no effect on economic activity. In particular, the simulations suggest that negative oil-price shocks create downward pressure on core inflation (through the expectations channel) that allows monetary policy to ease, thereby stimulating the economy. However, asymmetry in the real-wage catch-up term does appear to be a potentially significant and plausible explanation of less extreme characterisations of the observed nonlinear relationship between oil prices and macroeconomic activity. (23)

The implications of a delayed monetary policy response

Policymakers may have several reasons for delaying or moderating their responses to large increases in oil prices. First, analysis of historical experience suggests that it is difficult, if not impossible, to predict how sustained movements in oil prices might turn out to be. Second, there is considerable uncertainty about the effects of oil-price increases on core inflation. These factors may argue against responding to higher oil prices until they have been sustained for some time and their macroeconomic effects have become more apparent.

The simulations presented in this section are designed to illustrate what the possible benefits and potential dangers of delaying the policy response might be. We use the version of the model that allows oil price innovations to pass through into core inflation, and consider the implications of both symmetric and asymmetric real-wage catch up effects. For certain other cases -- in particular, for shocks that are very short-lived or for economies in which there is no risk of pass-through into core inflation -- delaying the policy response is the right thing to do.

It is widely recognised that the manner in which monetary policy responds to various shocks to the economy can have an important influence on inflation expectations. The main potential danger of delaying the response to an oil price increase stems from the possibility that delay may weaken the credibility of announced or perceived policy objectives and have an adverse effect on inflation expectations. Although modelling the evolution of policy credibility is a difficult task, analysis that abstracts from credibility issues can be very misleading.

For purposes of taking credibility issues into account, we compare simulations based on two alternative formulations of how delayed monetary policy reactions might affect the process for inflation expectations. Recall that MULTIMOD's base-case monetary policy reaction function is a forward-looking inflation-forecast-based rule under which the nominal short-term interest rate is adjusted (relative to an equilibrium level) in proportion to changes in the deviation of observed output from potential output and the deviation of forecast core inflation from an inflation target. Under this reaction function, adjustments to changes in the output gap and the inflation forecast occur with no delay. Moreover, expected inflation is modelled as partly backward-looking and partly forward-looking, where the forward-looking component is model consistent in that it depends on the structure of the model, including the monetary policy reaction function. In the first formulation of what delayed monetary policy reactions might im ply, we adopt the extreme assumption that delay has no effect on the inflation expectations process. In the second formulation, which is intended to illustrate the possible effects of an erosion of monetary policy credibility, we add an endogenous element to private agents' point estimate (perception) of the inflation target that enters the monetary policy reaction function. The inflation target perceived by private agents influences their expectations about future inflation, which in turn influences actual inflation outcomes.

The perceived inflation target is made endogenous in a simple and admittedly ad hoc manner. Policymakers lose credibility if they do not respond to the shock and the outcome for core inflation exceeds the inflation target by more than one-half percentage point. When this happens, beginning in the next year private agents revise their perception of the policymaker's target to be equal to the previous year's core inflation outcome. Private agents continue to base their perception of the policymaker's target on inflation outcomes until core inflation is returned to within one-half percentage point of the 'true' target. Once this occurs, private agents' perception of the target reverts to the true target. Because this model of endogenous policy credibility is ad hoc, it is important not to take the specific magnitudes of its effects too seriously; but the results provide a useful qualitative picture.

Chart 4 shows the dynamic adjustment paths for output, core inflation, and the policy interest rate under the assumption of symmetric real-income catch-up effects, while chart 5 describes the analogous adjustment paths under asymmetric real-income catch-up effects. Simulation results are provided for the cases of an immediate policy response, a delayed response with exogenous policy credibility, and a delayed response with endogenous policy credibility; in each case the shock is the persistent 50 per cent increase in oil prices (i.e., an increase that persists for two years and then erodes over the next three years, as described earlier). For the cases in which the policy response is delayed, the interest rate is held unchanged (relative to the baseline path) for one and a half years, after which it reverts to the path dictated by the monetary policy reaction function. (24)

The results of delaying the policy response vary somewhat between the United States and the Euro Area. In general, delaying the policy reaction -- that is, holding the short-term nominal interest rate constant and allowing inflation expectations to rise -- has expansionary effects on aggregate demand and GDP in the short run but leads to a much sharper interest-rate response by he third year, with subsequent contractionary effects on GDP. Interest rates are lowered as (forecast) inflation is brought under control, but both interest rates and output continue to cycle after oil prices revert (in year six) to the baseline level. The decline in interest rates - to levels below baseline in the simulations with exogenous credibility - reflects the decline in oil prices (beginning in year three) and the induced weakening of pressures on core inflation.

Table 5 reports summary statistics on the cumulative changes after ten years in output and the core price level. Comparisons of the entries in the first and second rows for each country indicate the cumulative effects of delaying the policy response when policy credibility does not suffer. The cumulative output foregone in these cases is relatively small in the United Sates but larger in the Euro Area, while the additional cumulative effect on the core price level ranges from 2.0 to 2.5 per cent. When credibility is endogenous, the cumulative output losses and price level increases are larger, as indicated by the differences between the first and third rows for each country.

The most striking implications of endogenous credibility, however, are not the differences apparent in table 5, but rather the interest rate implications shown in charts 4 and 5. Under endogenous credibility, the efforts to restore macroeconomic stability (i.e., to steer economies back to baseline), after delaying policy responses for a year and a half, result in short-term interest rates being pushed in the third year to levels about 2 percentage points above baseline in the United States and 1.5 percentage points in the Euro Area. (25) While sharp interest rate hikes might be regarded as successful in averting major cumulative costs in these hypothetical scenarios, in reality the scope for such an aggressive tightening of monetary policy is often constrained by political pressures. In the presence of such constraints, the costs of delay could be much larger than those summarised in table 5.

Comparative costs of potential policy errors

Given that the optimal policy response to an increase in oil prices depends so heavily on how private agents respond, what should monetary authorities assume when they set policy? Is there a danger that oil price increases will pass through into core inflation, or will oil price shocks simply result in temporary increases in head me inflation? Hooker's (1999) evidence for the United States, based on empirical analysis in a reduced-form Phillips curve framework, suggests that oil price increases did not affect core inflation during the 1980s and 1990s. However, policymakers may want to interpret this evidence with caution because the positive oil price shocks that occurred during the last two decades were very short-lived, and because any impacts on core inflation may have been mitigated by the response of policy.

It is broadly agreed that policymakers - in responding to oil prices increases, and in seeking to stabilise the economy more generally - should carefully monitor a wide range of indicators, and exploit the best analytic tools available, to try to reduce their uncertainties about economic behaviour, including the pass-through effects of inflationary shocks. But even with the best analysis available, policymakers must make decisions based on very incomplete information about macroeconomic relationships. Should they base their responses to increases in oil prices on the reduced-form Phillips curve evidence from the 1980s and 1990s; or should they respond as if there is potential for oil prices to influence core inflation? No matter what they assume, they are likely to be wrong to some degree. If the monetary authorities expect the inflationary consequences to be more persistent than turns out to be the case, then policy settings will initially be inappropriately tight, other things being equal. If, on the other hand, the authorities initially underestimate the strength of the pass-through effects into core inflation, then policy may not be tightened enough to stabilise inflation quickly.

Our purpose in this section is to illustrate that the macroeconomic costs of these two types of policy errors are of different orders of magnitude, and that the welfare-maximising strategy is for policymakers to base their responses on high-side estimates of the degree of passthrough into core inflation, other things being equal. We demonstrate this by using MULTIMOD to compare the effects on output and inflation of the persistent 50 per cent increase in oil prices under two possible structures for the economy combined with monetary policy reactions based, alternatively, on correct and incorrect information about those structures. The first structure has no core inflation consequences of the oil price increase. Under the second structure, the oil price shock passes through into core inflation, workers behave asymmetrically in only resisting the declines in their real wages that result from changes in oil prices, and the credibility of inflation objectives is eroded when core inflation deviates significantly f rom target in association with policy behaviour that does not provide an adequate response to the shock.

Four simulations are used to generate the estimated costs of making policy errors. In two of the simulations the policymaker correctly perceives the inflation process. In the remaining two simulations the monetary authorities misperceive the structure of the wage/price nexus for the first two years following the oil price shock, and policy during those years is based on a forecast of inflation that is generated using the incorrect model of the inflation process. The consequences of the policy errors are characterised by the additional macroeconomic variability that they cause, as measured by differences between the deviations of output and inflation from baseline in the simulations with policy errors minus the corresponding deviations from baseline in the simulations with the same structure of the economy and no policy errors. We define 'error 1' as the case in which policymakers assume no pass-through into core inflation when the true structure of the economy includes the asymmetric real-wage catch-up effect and credibility is endogenous. 'Error 2' is the case in which there is no pass-through into core inflation but policymakers base their reactions on the model of the economy that includes the asymmetric pass-through structure and endogenous credibility.

Table 6 presents several summary statistics that characterise the estimated costs of the errors. The characteristics of the experiments that are summarised are the additional variability that arise from the policy errors - i.e., the deviations from baseline under the policy error minus the deviations from baseline under the same structure of the economy with no policy error. The first two columns of the table report the additional cumulative changes in output and the core price level that arise from the errors. The third column of the table reports the loss that would arise under a quadratic loss function that equally weights the additional variability in real output and core inflation arising from the errors. (26)

The table shows that initially making the incorrect assumption that there are no core inflation effects (error 1) results in a permanent sacrifice of real output for the United States and the Euro Area. In these cases the slow response of policymakers implies that output must be permanently sacrificed to re-anchor inflation expectations at the target rate. Under the second error, the initial declines in real output that arise from policy being set too tight are fully recovered once policymakers (after two years) recognise their error. The price levels also exhibit more drift under the first policy error than they do under the second. And since both output and inflation show greater additional variability under the first error than under the second, the quadratic loss echoes the result that underestimating the strength of the passthrough effect is more costly than overestimating it. (27)

Concluding remarks

Oil price shocks posed major challenges for monetary policy during the 1970s. Subsequently, the price of oil exhibited fairly moderate variability during much of the 1980s and 1990s, but turned down sharply after the Asian financial crises erupted. Oil prices bottomed out at a 25-year low in February 1999, then tripled in the period through to early September 2000. Although the pronounced weakening of the global economy during 2001 has brought a reversal of much of the oil-price rise during 1999-2000 and alleviated the associated policy concerns, the recent episode generated renewed interest in analysing the expected macroeconomic consequences and risks associated with a rise in oil prices and the implications for how monetary policy should respond.

This paper has used MULTIMOD to address these issues. It has distinguished between the effects of temporary, more persistent, and permanent shocks and has suggested that even the effects of large permanent oil-price increases can be limited under forward-looking and well-chosen reactions by the monetary authorities. While the paper focuses on the effects of increases in oil prices, the framework can be applied symmetrically to study the implications of oil price declines.

The simulation analysis has developed several perspectives that are relevant for arriving at well-chosen monetary policy reactions and avoiding a repeat of the types of experiences that followed the oil shocks of the 1970s. Three perspectives deserve particular emphasis. First, experience during the 1980s and 1990s does not provide a valid basis for dismissing the risk that persistent oil-price increases will pass through into core inflation. Second, delay in responding to a persistent oil-price increase can have high macroeconomic costs if it leads to an erosion of monetary policy credibility. And third, in the face of significant uncertainties about behavioural relationships, monetary policymakers should interpret the data in a manner that errs in the direction of a more aggressive policy response to oil-price increases, other things being equal.

These conceptual perspectives do not imply that monetary policy should always respond immediately and/or aggressively to oil price increases that are expected (based on futures prices) to be persistent. But they do underscore the importance of carefully monitoring a wide range of economic indicators to watch for signs that oil-price increases may be threatening to pass into core inflation, of looking and listening for any indications that market participants might be beginning to doubt the credibility of monetary policy, and of exploiting the best analytic tools available to help narrow uncertainties about the nature and parameters of key behavioural relationships.

(*.) Research Department, International Monetary Fund. The views expressed in this paper are those of the authors and do not necessarily reflect those of the IMF. The authors are grateful to John Helliwell and Warwick McKibbin for helpful discussions.

NOTES

(1.) See Hooker (1999) and Hamilton (2000) for references.

(2.) Hamilton's (2000) approach to characterising the relationship between oil price changes and GDP growth is flexible enough to test a broad class of nonlinear specifications but does not have the power to distinguish between the different form of the nonlinearity proposed in Mork (1989), Lee and others (1995) and Hamilton (1996). Hamilton (2000) also demonstrates that the data support the hypothesis that oil prices have a linear (symmetric) effect on economic activity when the analysis is conducted with an instrumental variables regression in which identifiable exogenous disruptions in world petroleum supplies are used as instruments; this alternative interpretation appeals to the argument that the distribution of historically-observed exogenous shocks is asymmetric.

(3.) The methodology of their paper elicited some questions from participants on the Brookings Panel to which it was presented, but participants generally accepted the conclusion that the output declines following oil price shocks had come mainly from monetary policy responses.

(4.) This list abstracts from induced effects on asset values and their implications for aggregate demand and supply, and also from the effects of any induced changes in fiscal position or fiscal policies.

(5.) Laxton and others (1998) describe the Mark Ill version; see also Isard (2000). The version of Multimod used to conduct the simulations for this paper is an extended Mark III version referred to as Mark IIIA. The major revisions included in Mark Illa include: the incorporation of a Euro Area block; new base-case specifications of the behaviour of monetary and fiscal policy; and a re-coding of the model that more easily permits solutions to the model in which countries choose different steady-state rates of inflation. The Mark IIIA version contains six industrial country blocks and two developing country blocks. The six industrial country blocks, the United Sates, the Euro Area, Japan, the United Kingdom, Canada and the Other industrial countries have an identical structure with parameter values that are allowed in some cases to be country specific. The developing country sectors of MULTIMOD contain highly oversimplified descriptions of macroeconomic behaviour that serve to ensure the global consistency o f MULTIMOD simulations but provide only a minimal characterisation of the channels through which changes in oil prices affect the developing economies.

(6.) MULTIMOD is by no means completely immune from Lucas Critique problems. The Phillips curve, for example, is a reduced-form equation, and there is always the possibility that a major change in the pattern of monetary policy behaviour could lead to significant changes in the nature of wage and price contracts and the dynamics of inflation expectations.

(7.) For a discussion of the potential benefits of IFB rules see Isard, Laxton, and Eliasson (1999), Drew and Hunt (2000), and Clark, Laxton, and Rose (2001).

(8.) Helliwell (1986, 1987) and McKibbin and Sachs (1991) consider a more general production function that allows for a more explicit role for inputs of primary and intermediate products.

(9.) IMF Research Department (2000).

(10.) We would note that the degree of uncertainty about the level of potential output that could arise from even quite large oil price shocks seems very small relative to the magnitued of other uncertainties about potential output.

(11.) Unlike many macroeconometric models, however, MULTIMOD Mark IIIA's reduced-form Phillips curves are nonlinear with respect to labour market disequilibria. This feature allows for the possibility that large policy errors can have first-order welfare implications.

(12.) Equation 1 has been estimated for each of MULTIMOD's major industrial countries/blocks as part of an unobserved components model that also includes equations for the deterministic-NAIRU, the NAIRU, and an Okun's Law relationship between the output gap and the unemployment gap. The estimation is done using the Kalman filter and a constrained-maximum-likelihood procedure. Equations 2 and 3 were estimated with OLS. More details regarding the model and its estimation can be found in Hunt and others (2002).

(13.) A 50 per cent increase in the price of oil is equivalent to an increase in [[phi].sup.POIL] of 40.55 because [[phi].sup.POIL] is defined to be 100 times the first-difference of the log of the oil price. For changes as large as 50% there can be substantial differences between changes in logs and per cent changes.

(14.) The reaction function sets the short-term nominal interest rate equal to an 'equilibrium' nominal interest rate plus 0.5 times the output gap plus 1.0 times the deviation from target of the one-year ahead forecast for core inflation. The equilibrium nominal interest rate is defined as an equilibrium real interest rate plus the expected rate of inflation (as given by equation 3).

(15.) See, for example, Isard, Laxton, and Eliasson (1999).

(16.) The United States and the Euro Area exhibit the largest overall pass-throughs of oil-price shocks into core inflation, with the Euro Area experiencing relatively strong transmission through the expectations channel and relatively weak real-income catch-up effects.

(17.) IMF Research Department (2000).

(18.) Hooker (1999).

(19.) Accordingly, the simulation results presented in table 3 may overstate the differences that would have arisen under the two alternative wage/price structures in response to the very temporary innovations to oil prices that occurred during the late 1980s and 1990s.

(20.) A Monte Carlo experiment could be set up to test this hypothesis more formally. Artificial data could be generated under the two alternative model structures allowing the oil price shocks in the experiment to differ in their persistence. Reduced-form Philips curves could then be estimated on the artificial data, testing whether the persistence of the oil shocks mattered for the identification of the true model.

(21.) The empirical work undertaken in estimating MULTIMOD's wage/price nexus was unable to identify an asymmetric relationship, and the base-case version of MULTIMOD Mark IIIA contains a symmetric specification. The failure to identify an asymmetry is not surprising, given that an identical parsimonious specification was estimated for each country/block on annual data with few observations of persistent declines in oil prices.

(22.) Simulations of persistent shocks (i.e., oil price changes that dissipate over a 5-year period) generate anomalous results under our simple characterisation of the asymmetry in wage/price behaviour. In particular, simulations of persistent increases and declines in oil prices suggest an even greater asymmetry in the first few years of the shock than simulations of permanent shocks, with persistent negative shocks to oil prices leading to declines in output after several years. The latter anomaly arises because the asymmetry in real-income catch-up effects has workers trying to lock in the real wage gains that occur when oil prices initially decline. As oil prices recover, workers resist the decline in real incomes from their recently improved levels and monetary policy then needs to tighten to combat the inflationary effect.

(23.) As noted above, another plausible explanation of the observed nonlinearity is Hamilton's (2000) suggestion that the distribution of the exogenous component of historically-observed oil-price changes has been asymmetric, with most exogenous changes in oil prices consisting of price increases associated with war-induced petroleum supply disruptions.

(24.) More precisely, the policy setting in the second year is the average of what would emerge from the case with a one-year delay and the case with a two-year delay, with the interest rate reverting in the third year to the path dictated by the policy reaction function.

(25.) Short-term interest rates are pushed up slightly more in the case of a symmetric real-wage catch-up effect than in the asymmetric case, but short-rates also decline somewhat more gradually in the latter case. Long-term interest rates, which reflect projected short rates, tend to increase somewhat more in the asymmetric case.

(26.) The loss is calculated as the sum, over the first ten annual observations, of the square of the difference between the two deviations of real GDP from baseline, plus the square of the difference between the two deviations of core inflation from target (baseline).

(27.) Applying a 5 per cent annual discount factor or including interest rate variability with a weight of 0.3 in the loss calculation does not alter the ranking

REFERENCES

Bernanke, B., Gertler, M. and Watson, M. (1997), 'Systematic monetary policy and the effects of oil price shocks', Brookings Papers on Economic Activity, I, pp. 91-157.

Cashin, P., McDermott, J. and Scott, A. (1999), 'Booms and slumps in world commodity prices', IMF Working Paper WP/99/l55.

Clark, P., Laxton, D. and Rose, D. (2001), 'An evaluation of alternative monetary policy rules in a model with capacity constraints', Journal of Money Credit and Banking 33, I, February, pp. 42-64.

Drew, A. and Hunt, B. (2000), 'Efficient policy rules and the implications of potential output uncertainty', Journal of Economics and Business, 52, 1/2, January/April, pp. 143-60.

Hamilton, J. (1983), 'Oil and the macroeconomy since World War II., Journal of Political Economy. 91, pp. 228-48.

--(1996), 'This is what happened to the oil price-macroeconomy relationship'. Journal of Monetary Economics, 38, pp. 215-20.

-- (2000), 'What is an oil shock?' NBER Working Paper 7755, June.

Helliwell, J. (1986), 'Supply side in the OECD's macroeconomic model', OECD Economic Studies, 6, Spring, pp. 75-131.

--(1987), 'Supply oriented macroeconomics: the MACE model of Canada', Economic Modeling, July, pp. 318-340.

Hooker, M. (1996), 'What happened to the oil price-macroeconomy relationship', Federal Reserve Board (FEDS) Working Paper No. 56.

--(1999), 'Are oil shocks inflationary? Asymmetric and nonlinear specifications versus changes in regime', Working Paper, Federal Reserve Board, December.

Hunt, B., Isard, P. and Laxton, D. (2001), 'The macroeconomic effects of higher oil prices', IMF Working Paper WP/01/14.

IMF Research Department (2000), 'The impact of higher oil prices on the global economy', Washington, International Monetary Fund, available at www.imf.org/external/pubs/ft/oil/2000/index.htm.

Isard, P. (2000), 'The role of MULTIMOD in the IMF's policy analysis', IMF Policy Discussion Paper PDP/00/5.

Isard, P., Laxton, D. and Eliasson, A. (1999), 'Simple monetary policy rules under model uncertainty', in Isard, P., Razin, A. and Rose, A.K. (eds), International Finance and Financial Crises: Essays in Honor of Robert P. Flood, Jr., Washington and Boston, International Monetary Fund and Kluwer, pp. 207-50.

Laxton, D., Isard, P., Faruqee, H., Prasad, E. and Turtelboom, B. (1998), 'MULTIMOD Mark III: the core dynamic and steady-state models', IMF Occasional Paper No. 164.

Lee, K., Ni, S. and Ratti, R. (1995), 'Oil shocks and the macroeconomy: the role of price variability', Energy Journal, 18, pp. 39-56.

Lucas, R. E. Jr. (1976), 'Econometric policy evaluation: a critique', in Brunner, K. and Meltzer, A.H. (eds), The Phillips Curve and Labor Markets, Carnegie-Rochester Conference Series on Public Policy, I, Amsterdam, North Holland.

McKibbin, W. and Sachs, J. (1991), Global Linkages: Macroeconomic Interdependence and Cooperation in the Wand Economy, Washington D.C., The Brookings Institution.

Mork, K. (1989), 'Oil and the macroeconomy when prices go up and down: an extension of Hamilton's results', Journal of Political Economy, 97, pp. 740-4.

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

[Graph omitted]

Table 1

MULTIMOD base-case parameters

 10[[delta].sub.3] [[delta].sub.2]

Average 0.22 0.58
United States 0.31 0.58
Euro Area 0.33 0.44
Japan 0.14 0.73
United Kingdom 0.15 0.69
Canada 0.20 0.61
Other industrial countries 0.18 0.41

 [[delta].sub.1] [psi] [alpha] [OMEGA]

Average 0.08 0.54 0.26 0.57
United States 0.08 0.51 0.35 0.53
Euro Area 0.12 0.51 0.12 0.58
Japan 0.06 0.59 0.09 0.60
United Kingdom 0.11 0.58 0.42 0.60
Canada 0.06 0.51 0.16 0.50
Other industrial countries 0.08 0.55 0.42 0.60

 [lambda]

Average 0.48
United States 0.48
Euro Area 0.60
Japan 0.31
United Kingdom 0.34
Canada 0.41
Other industrial countries 0.74
Table 2

Contemporaneous direct effects on annual CPI inflation of a 50 per cent
increase in the price of oil (Deviations from control in percentage
points)


Average 0.9
United States 1.3
Euro Area 1.3
Japan 0.6
United Kingdom 0.6
Canada 0.8
Other Industrial Countries 0.7

Note: The measures of inflation in the model in the main text
[[[pi].sup.CPI], [[pi].sup.M], [[pi].sup.POIL], [[pi].sup.c]] are
defined to be 100 times the first-difference of the log of each
variable. This measure of inflation will only be approximately
equal to the percent change in the series when the change is fairly
small.
Table 3.

Temporary versus persistent increases oil prices (shock minus control)

Variable Shock Pass-through Year
 to core 1 2
 inflation

The United States

GDP Temporary No 0.0 0.0
 Yes 0.0 0.0
 Persistent No 0.0 0.0
 Yes -0.2 -0.4

CPI

Inflation Temporary No 1.3 -0.9
 Yes 1.3 -0.5
 Persistent No 1.3 0.4
 Yes 1.4 0.9

Core

Inflation Temporary No 0.0 0.0
 Yes 0.0 0.5
 Persistent No 0.0 0.0
 Yes 0.2 0.7

The Euro Area

GDP Temporary No 0.0 -0.1
 Yes 0.0 0.0
 Persistent No 0.0 -0.1
 Yes -0.1 -0.4

CPI

Inflation Temporary No 1.4 -0.8
 Yes 1.4 -0.7
 Persistent No 1.5 0.6
 Yes 1.6 1.1

Core

Inflation Temporary No 0.0 0.0
 Yes -0.1 0.2
 Persistent No 0.0 0.0
 Yes 0.2 0.6

Variable Year
 3 4 5


The United States

GDP 0.0 0.0 0.0
 0.1 0.1 0.1
 0.0 0.0 0.0
 -0.3 0.0 0.1

CPI

Inflation -0.3 -0.1 0.0
 -0.3 -0.2 -0.1
 -0.1 -0.3 -0.4
 0.4 0.1 -0.2

Core

Inflation 0.0 0.0 0.0
 -0.1 -0.2 -0.0
 0.0 0.0 0.0
 0.5 0.2 0.0

The Euro Area

GDP 0.0 0.1 0.0
 0.1 0.2 0.1
 -0.1 0.0 0.0
 -0.3 0.0 0.2

CPI

Inflation -0.4 -0.2 -0.1
 -0.4 -0.3 -0.2
 0.0 -0.3 -0.5
 0.5 0.1 -0.3

Core

Inflation 0.0 0.0 0.0
 -0.1 -0.2 -0.1
 0.0 0.0 0.0
 0.5 0.2 0.0
Table 4.

The effects of permanent oil price shocks with asymmetry in the
real-wage catch-up effect (shock minus control)

Variable Shock Year
 1 2 3 4

The United States

GDP Positive -0.3 -0.5 -0.5 -0.3
 Negative 0.1 0.2 0.2 0.1

CPI

inflation Positive 1.4 0.9 0.6 0.4
 Negative -1.4 -0.6 -0.3 -0.1
Core

inflation Positive 0.1 0.7 0.5 0.3
 Negative -0.1 -0.2 -0.2 -0.1

Short-term

interest rate Positive 1.1 1.0 0.7 0.4
 Negative -0.4 -0.5 -0.2 -0.1

The Euro Area

GDP Positive -0.2 -0.6 -0.6 -0.4
 Negative 0.1 0.3 0.3 0.2

CPI

inflation Positive 1.6 1.1 0.8 0.5
 Negative -1.5 -0.9 -0.6 -0.3

Core

inflation Positive 0.2 0.6 0.5 0.3
 Negative -0.2 -0.4 -0.3 -0.2

Short-term

interest rate Positive 1.2 1.2 0.8 0.5
 Negative -0.8 -0.9 -0.5 -0.2

Variable Year
 5

The United States

GDP -0.2
 0.0

CPI

inflation 0.2
 0.0
Core

inflation 0.1
 0.0

Short-term

interest rate 0.2
 0.0

The Euro Area

GDP -0.2
 0.1

CPI

inflation 0.3
 -0.2

Core

inflation 0.1
 0.0

Short-term

interest rate 0.2
 0.0
Table 5.

Summary statistics from delayed policy response simulations (percentage
points)

 Cumulative change after ten years
 RealGDP Core price level
 Symmetric Asymmetric Symmetric Asymmetric
 real-wage real-wage real-wage real-wage
 catch-up catch-up catch-up catch-up

United States
Immediate
 response 0.2 -0.6 0.4 0.8
Delayed
 response 0.1 -0.8 2.8 3.1
Delayed
 response
 erodes
 credibility -0.3 -1.3 4.2 4.5
Euro Area
Immediate
 response 0.7 0.2 0.2 0.3
Delayed
 response 0.3 0.0 2.2 2.4
Delayed
 response
 erodes
 credibility 0.2 -0.1 2.8 2.9
Table 6.

Costs of possible policy errors

Error 1 -- erroneously believe no core inflation effects

Error 2 -- erroneously believe core inflation effects

 Cumulative change after ten years
 (Percentage points) Quadratic loss
 (Inflation and
 Real GDP Core price level output variance)

United States

Error 1 -0.4 3.0 5.4
Error 2 0.1 -1.9 0.8
Euro Area
Error 1 -0.3 2.0 1.8
Error 2 0.1 -1.8 0.9