Oil Prices and Spanish Interest Rates

Oil Prices and Spanish Interest Rates Rebeca Jiménez-Rodríguez and Marcelo Sánchez # This paper studies the impact of oil shocks on Spanish short-term interest rates, focusing on the period since Spain s EU accession in 1986. Over this period, we find the following results. A linear model outperforms leading non-linear representations. In response to the shock, real short-term interest rates tend to rise, while nominal interest rates increase only gradually. The share of real interest rate variability explained by oil prices is around 10%. Historical decompositions show that oil prices impacted interest rates in periods of high oil prices, such as that around the 1990 spike and in some instances since 1999. Counterfactual analysis indicates that the increase in interest rates following an oil price shock has contributed to reducing inflation, while also making the economic slowdown more marked. The fraction of macroeconomic variability induced by the oil shock indirectly via monetary policy changes appears to be rather modest. The main results were not robust to starting the estimation in 1970, in which case we find oil prices have non-linear impacts, real interest rates are little affected by oil shocks and monetary policy appears to be accommodative. Keywords: Business cycles; Oil price shock; Spain; Interest rates; Counterfactuals JEL classification: E32; Q43 Department of Economics. University of Salamanca. Campus Miguel de Unamuno. E-37007. Salamanca. Spain. Tel.: +34 923 29 46 40 (ext. 35 14). Fax: +34 923 29 46 86. E-mail: rebeca.jimenez@usal.es # Corresponding author. European Central Bank. Kaiserstrasse 29. D-60311. Frankfurt am Main. Germany. Tel.: +49 69 1344 6531; fax: +49 69 1344 6353. E-mail address: marcelo.sanchez@ecb.int. 1

1. Introduction There is a large literature on the macroeconomic consequences of oil price shocks, which largely focuses on the latter s impact on domestic prices and especially real output. Given that crude oil is a basic input to production, the theory normally predicts that supply-side consequences of oil price hikes include a contraction in overall economic activity and inflationary pressures. In addition, aggregate demand is expected to fall in oil importing countries, and go up in oil exporting countries. Existing empirical work has by and large confirmed the results found in the theoretical literature. 1 The initial studies for the US identified a linear negative link between oil prices and real activity. It was eventually found that by the mid-1980s such linear relationship began to lose significance. The three leading non-linear approaches have been developed by Mork (1989), Lee et al. (1995), and Hamilton (1996) with the aim of re-establishing the negative relationship between increases in oil prices and real output developments. 2 Mork's (1989) study found that the real effects of oil price increases are different from those of decreases, with oil price decreases not having a statistically significant impact on US economic activity. Two other non-linear models were later developed in the literature, namely: the scaled specification (Lee et al., 1995), which takes the volatility of oil prices into account; and the net specification (Hamilton, 1996), which considers the amount by which oil prices have gone up over the last year. The impact of oil prices on interest rates has received comparatively less research attention. One important exception is the debate between Bernanke et al. (1997) and Hamilton and Herrera (2004) about the relative role played by oil shocks and monetary policy in US business cycle fluctuations. The former study suggests that monetary policy could be used to undo any recessionary consequences of an oil price shock. Hamilton and Herrera (2004) emphasise that the size of the effect that Bernanke et al. attribute to oil shocks is substantially smaller than the one that results from choosing an optimal lag length. The present paper extends the empirical work on oil price impacts by examining the case of Spanish interest rates. We focus on the period since Spain s EU accession in 1986. This topic has not been analysed to date in the relatively small number of studies covering Spain in the related literature. It is however potentially very interesting since Spain is an economy that, over the period under study, both exhibits considerable oil import dependence thus being expected to be adversely affected by oil shocks - and has been found to play a considerable role in synthetic euro area developments. 3 The approach to model estimation pursued here follows Jiménez-Rodríguez and Sánchez's (2005) use of the linear and three leading non-linear methods described above. In doing so, we deviate from the focus in much of the related literature on real output developments. Here we concentrate on short-term interest rates, although also 1 For a recent survey of the literature on the US, see Hamilton (2006). For evidence on countries outside the US, see Mork et al. (1994), Cuñado and Pérez de Gracia (2003), and Jiménez-Rodríguez and Sánchez (2005). 2 For other references relating to the structural break in the US case towards the mid-1980s, see Hooker (1996) and Hamilton (1996). 3 On the latter, see Mojon and Peersman (2003). 2

exploring the latter s structural relationship with both inflation and economic activity. Our analysis combines impulse responses (including counterfactual experiments), variance decompositions and historical decompositions the latter showing the contribution of oil prices to interest rate fluctuations over time. Our paper differs from previous analyses of macro impacts of oil prices in Spain in both scope and methodology. De Miguel et al. (2003) use an extended real business cycle model to analyse the real output impact of the shock, estimating that a one standard deviation shock to real oil prices (amounting to about a 20% hike) induces a fall in real output of some 0.5% in the short run. Barrell and Pomerantz (2004), using NiGEM international simulations as from 2005, compute the real output impact of an oil price shock in Spain, showing that this impact is rather modest over the first year and that it gains in strength over the following three years. Cuñado and Pérez de Gracia (2003), while focusing on other European economies in their impulse response analysis, report some results for Spain. In particular, the negative correlation between oil prices and industrial production is found to drop somewhat after the mid-1980s. The rest of the paper is organised as follows. Section 2 describes the methodology. Section 3 presents the empirical results. Section 4 concludes. 2. Methodology The first step consists of estimating a reduced form as given by a vector autoregression model of order p, or simply, VAR(p). We can write this system as yt = Ax t + ε t where y t is an n 1 vector of endogenous variables, x t is an np 1 vector grouping all lagged terms of y t up to order p, A is an n np rectangular matrix, and ε t is an n 1 vector of white noise processes with variancecovariance matrix Σ. The optimal lag length for the VAR is determined by using the likelihood ratio test. We consider the following set of endogenous variables: real GDP, real effective exchange rate (REER), real oil price, real wage (CPI deflated), CPI inflation, and real short- and long-term interest rates. Some variables (real GDP, REER, real oil price and real wage) are expressed in logs, while the remaining ones are simply defined in levels. Oil prices, real GDP and inflation are included since they are the main variables of interest of this study. The remaining variables in the model are added in order to capture the most important transmission channels through which oil prices may affect economic activity indirectly, in particular by inducing changes in monetary policy. Those channels include a variety of demand- and supply-side effects of oil prices operating via exchange rates, financial variables and the labour market. The data used are quarterly and run from 1970:1 through 2006:1. Our measure of real oil prices is defined as the ratio of the price of an internationally traded variety of crude (UK Brent) in US dollars to the US Producer Price Index (both from IMF's International Financial Statistics - henceforth IFS). 4 Regarding Spanish data, real 4 Oil prices are used directly in the linear approach to VAR estimation, and are transformed - in ways discussed below - for their use in non-linear models. The choice of a readily available Producer Price Index for the denominator of our oil price measure helps reduce the possibility that the identified oil 3

GDP data is from the BIS; CPI from the OECD's Main Economic Indicators (henceforth MEI); the short-term interest rate from the BIS; long-term interest rates from OECD's Economic Outlook; nominal (hourly) wages from IFS; and REER (based on CPI) from MEI. Turning to the study of the stochastic properties of the series considered here, we assess their order of integration by means of unit root tests. Specifically, we perform the DFGLS and P T tests of Elliott et al. (1996), and the DFGLS u and Q T tests of Elliott (1999), as well as the Augmented Dickey-Fuller (ADF) test. Results of these formal tests are summarised in Table 1. We shall use a model in which the vector of endogenous variables is given by the first log-differences of four variables (real GDP, REER, real oil price, and real wage), along with the levels of inflation, and short and long-term real interest rates. From Table 1, the growth rates of real GDP, real oil price, REER and the real wage appear to be stationary. In the absence of model misspecification, the remaining variables are the levels of real interest rates and inflation as this enhances economic interpretability. 5 In light of possible nonstationarity of inflation and the real interest rate measures, we have also estimated all models with these variables in first differences. The results are however not substantially affected compared to the ones reported in this paper. Our baseline approach to identification of oil price shocks consists of a Cholesky decomposition in which the real oil price is put first in the recursive ordering for the variables in the system. The oil price variable is thus ranked as the most exogenous variable, being allowed to have an immediate impact on the remaining macroeconomic variables. As a robustness check, we also report in subsection 3.5 impulse response results from an alternative identification to one presented earlier, imposing short-run restrictions. As mentioned earlier, we focus on the period since Spain s accession to EU membership in 1986. Given our focus on interest rates, the choice of 1986 appears to be a reasonable choice. Indeed, the existing literature suggests that, by then, a switch in the monetary policy regime was well underway, with the role of the interest rate as an instrument gaining importance at the expense of monetary intermediate targets (see e.g. Ayuso and Escrivá, 1998, and Camarero et al., 2002). Along these lines, López (2002) estimates a Taylor rule for Spain over the period 1984-1998. We also report results from a full sample period starting in 1970. For both of our main sample periods (namely, 1986:1-2006:1 and 1970:2-2006:1), we aim at checking whether our results change when we shorten the estimation sample to the period before Spain joins the European Monetary Union (EMU) in 1999. However, depending on the optimal lag length for estimation over the period since 1986, we might not pursue such robustness check if the degrees of freedom were to become too small. 6 shock mistakenly captures an inflation uncertainty component. The GDP deflator has the advantage of being more comprehensive, but it is also reported with a long lag. 5 In connection with this, Alonso et al. (2000) find that real short-term interest rates have information for inflation. This does not mean that we lose track of nominal interest rates, given that they can be traced from developments in real rates and inflation. 6 Other periods in which relevant changes to monetary policy took place include the enactment of central bank autonomy in 1994 and the introduction of inflation targets in January 1995. 4

In addition to linear vector autoregressions (VAR), we estimate the three leading nonlinear approaches, namely, Mork's (1989) asymmetric model, Lee et al.'s (1995) scaled model, and Hamilton's (1996) net oil model. Such non-linear models have proved useful to capture the macroeconomic effects of oil prices, often beating linear models in terms of stability when estimated over longer sample periods. The asymmetric specification allocates positive realisations of the rate of change in the oil + t price to variable o, and negative such realisations to o. In the scaled approach, the relevant oil variable - standing for "scaled oil price increases" - is ^ ^ SOPI t = max t t 0, e / h, where e^ t and h^ t are, respectively, the error and the conditional variance of the rate of change in oil prices from a AR(4)-GARCH(1,1) representation. Finally, the net oil model uses the "net oil price increase" variable, defined as the amount by which oil prices (in logs), p t, exceed - if at all - the maximum value over the previous 4 periods (quarters); that is, NOPI max 0, p max p, p, p p. 7 ( { }) t = t t 1 t 2 t 3, t 4 3. Empirical results This section reports all the empirical results of this paper. These results are presented with a focus on the sample starting in 1986. Two alternative estimation samples a full sample one and, in some cases, one covering the period 1970-1998 - are also discussed. The reason for not reporting estimations over 1986-1998 is that the optimal lag for the model estimated since 1986 is six, which would leave us with too small degrees of freedom if we were to shorten the end of the sample to 1998. Of course, this situation is rather unfortunate since it would be interesting to investigate whether joining the EMU has made a difference for Spain. More concretely, the next subsections include the following results. First, we report a number of preliminary tests for significance and model selection. Second, we describe our results for accumulated responses of real GDP growth, inflation and real shortterm interest rates to oil price shocks. Third, we discuss our results for the corresponding variance decompositions. When describing our results, we focus on the approach (linear or non-linear) that performs best for each sample. Fourth, we report historical decompositions. Finally, we present results from an alternative model. The latter s reduced form is of the near-var type (with oil prices not being affected by lags of the remaining variables), while the identification approach used relies on short-run restrictions. This alternative modelling approach is used for robustness and counterfactual analysis. t 3.1. Testing for significance and model selection In this subsection we investigate the significance of the relationship between oil prices and the other variables of the model. We test for the significance of the oil price variables under consideration for the VAR system as a whole, using the null 7 We similarly construct scaled and net oil price decrease variables, that is, SOPD t and NOPD t, respectively. As we show in the next section, neither of the two is however found to have statistically significant macroeconomic effects in most of the cases. 5

hypothesis that all of the oil price coefficients are jointly zero in all equations of the system but its own equation (see Table 2). The likelihood ratio test is informative in that oil prices, in addition to their direct effect on real GDP and inflation, could well impact the latter two variables through the rest of the system. We find that the oil price variable in the linear model, as well as the positive changes in the asymmetric, scaled and net models, are all significant at the 5% level for the system for all three sample periods considered. Moreover, the negative changes in the asymmetric, scaled, and net models enter significantly in all models estimated for the period since 1986. In contrast, these negative changes are not significant in any of the two periods starting in 1970, except for the corresponding net specifications. The price decrease variable is subsequently eliminated from those non-linear specifications in which it is not significant. For model selection purposes, we look at two selection criteria as given by the Akaike information criterion (AIC) and Schwarz Bayesian information criterion (BIC). Table 3 reports the AIC and BIC obtained from each econometric specification. 8 On the basis of these two criteria, we conclude that the linear model turns out to be the bestperforming for the post-1986 period, while the scaled specification is the preferred model for the full sample and the period 1970-1998. 9 Interestingly, this stands in contrast with the literature for the US, which finds that the linear model breaks down in the mid-1980s and is thereafter outperformed by non-linear specifications. 10 We now turn to the description of the main results of the paper, concentrating on the preferred specifications for each sample. 3.2 Accumulated impulse responses This subsection contains a discussion of the impact of oil price shocks on real GDP, inflation and short-term interest rates, with a focus on the preferred model. Table 4 reports accumulated responses to a 100% oil price shock for the best-performing models. The latter are given by linear specification for the period since 1986 and the scaled approach for the full sample and the period 1976-1998. As described in section 2, the models used here include in the vector of endogenous variables the growth rates of three variables (real GDP, REER, and real wage), the real oil price variable, along with the levels of inflation, and short and long-term real interest rates. In line with Spain being a net oil importing economy, a qualitatively robust impulse response finding is that unexpected hikes in oil prices reduce real output and increase inflation. Focusing on the period since 1986, the impact of the oil shock on real GDP growth is an order of magnitude bigger than the inflationary pressures arising from 8 The optimal lag length was found to be six in all models for Spain, except those estimated over the period 1970-1998 in which case the optimal lag is four. 9 To our knowledge, the only previous study testing for asymmetric effects of oil prices in Spain is Cuñado and Pérez de Gracia (2003). They find that such effects are significant using a model for industrial production, inflation and oil prices, which they estimate using quarterly data for 1960-1999. 10 The full result concerning model selection for the full sample and the period 1970-1998 suggests that it is important to consider not just whether oil prices increase or decline (and by how much), but also the environment in which the movements take place. An oil shock in a stable price environment is likely to have larger economic consequences than one in a volatile price environment. In this regard, the scaled model more specifically highlights the importance of controlling for the time-varying conditional variab ility of oil price shocks. 6

the disturbance. For instance, after the second year the oil shock induces a reduction of 4.2% in real GDP while the inflationary effect is of only 1.2%. The impact on real GDP is larger in the full sample case in absolute terms. Since 1986, inflationary pressures appear to be larger than in the full sample case after two years, but they are relatively smaller at the end of both the first and third years. The finding that, as a result of an oil hike, Spanish economic activity contracts strongly is broadly consistent with the findings reported in Barrell and Pomerantz (2004) and De Miguel et al. (2006). 11 Finally, we also report the short-term interest rate responses to the exogenous oil price hike. For the post-1986 period, the real short-term interest rate is found to increase after the first and third year, while it is below baseline at the end of the second year. The corresponding nominal rate is above the baseline by 170 basis points at the end of both the first and second years, then rising further to 200 basis points at the end of the third year. The results for the full sample and the period 1970-1998 stand in sharp contrast to the ones just described. We find that real short-term interest rates decline, except at the end of the first year for the period 1970-1998 for which no change is reported. Both in the full sample case and the period 1970-1998, nominal interest rates are found to increase after the first year but thereafter fall after the second and third years. We will investigate the behaviour of interest rates further in subsection 3.5, in which identification of monetary policy shocks will allow us to draw monetary policy implications from impulse response results. 3.3. Variance decompositions Table 5 reports the variance decompositions of real GDP and inflation due to oil price shocks under the preferred specification for the period since 1986, the full sample and the period 1970-1998. The fraction of Spanish real output s forecast variance explained by oil prices is high, especially after the second and third years. These results happen to be very robust across estimation periods. The result that Spain s share of real output variability explained by oil prices tends to be rather high is consistent with De Miguel et al. s (2006) finding that a higher share of the empirical real output variability is explained by their theoretical model to be associated with oil price shocks. Turning to the share of inflation variability attributed to oil prices, it is lower than that of real output variability, especially in the two estimation periods starting in 1970. For the period since 1986, the share of inflation variability explained by oil prices at the end of the first year share is actually above the corresponding one for real GDP growth. Finally, the share of real short-term interest rate variability explained by oil prices is around 10% in the period since 1986. This value is found to lie well above that displayed over the entire sample and the period 1970-1998. 11 Granger-causality results in Cuñado and Pérez de Gracia (2003) point to the lack of significance in the lagged macroeconomic effects of oil prices in Spain. 7

3.4. Historical decompositions Here, we perform historical decompositions showing the contribution of oil prices to short-term interest rates over time. Figure 1 assesses the role of oil prices in the determination of real interest rates for the period since 1986 (top panel) and for the full sample (lower panel). Figure 2 displays the corresponding historical decompositions for nominal interest rates. In all these Figures, we report the actual series and the contribution of oil price shocks to the forecast, both of them in percentage deviations from a 6-quarter-ahead base projection. 12 Unexpected developments in oil prices appear to play a limited role in driving Spanish real short-term interest rates over time (see Figure 1). Using results since 1986, oil prices also appear to have contributed to movements in real rates in a muted way, with their impact being spread out over the entire estimation period. The contribution to a real interest rate hike is sizeable in one quarter (namely, 1991:1) at the end of the oil price spike following the 1990 Iraqi war. Another episode of higher oil prices where we observe an impact on real interest rate movements is in 1999-2001, with real rates reacting by moving down in 1999 and thereafter on the way up. When we look at full sample results, effects on real rates are visible in 1974-1975 and less so around 1990. [Insert Figure 1 around here] Oil price shocks are found to have exerted some influence on nominal short-term interest rates (see Figure 2). Our results for the sample starting in 1986 show that oil prices impacted interest rates in periods of high oil prices, such as that around the 1990 spike and more recently in 1999 and 2005-2006. The finding that oil shocks have driven nominal interest rates up in 1999 stands in contrast with the downward pressure on real interest rates displayed in Figure 1. This gap is explained by the rather quick and substantial inflationary effect taking place in that year. Employing full sample results, we detect an impact on nominal rates for 1974-1975, and to a smaller degree, around 1990. Overall, one rather robust and surprising - result is that no perceptible change in interest rates (either real or nominal) is found during the major oil shock of the late 1970s and early 1980s. [Insert Figure 2 around here] 3.5. Robustness and counterfactuals As a robustness check, here we report the impulse response results from an alternative VAR approach to one presented earlier. This alternative method relies on short-run restrictions for identification, while the reduced form of the model is estimated as a near-var system in which all lagged variables except oil prices are excluded from having any impact on the oil prices themselves. With regard to the short-run restrictions used, we draw from the identified VAR literature stemming from Cushman and Zha (1997), and Kim and Roubini (2000), while adapting it to the 12 Historical decompositions reported in the top panel of Figures 1 and 2 start in 1989:1, while those in the corresponding lower panels start in 1972:3. We also computed decompositions using different conditional forecasts, which tended to broadly corroborate our substantive results. 8

current environment. In doing so, we likewise combine a recursive block with some non-recursive exclusion contemporaneous restrictions. We set up a recursive block starting from oil prices which remain the most exogenous variable, followed by real GDP growth, inflation and real wage growth, in that order. We then postulate a monetary policy reaction function in which the real short-term interest rate is allowed to be immediately affected by all other variables but two: real output growth, which is assumed to be observable only with a lag, 13 and the real long-term interest rate which is rather allowed to be responsive to short-term rate fluctuations. Furthermore, the real long-term interest rate is allowed to react directly to all other variables except REER, which can only exert a contemporaneous impact on the former via the real short-term interest rate. Finally, REER may respond immediately to all shocks in the model. In particular, this implies that we allow the short-term interest rate to interact contemporaneously with the exchange rate, as in the identification approaches of Smets (1997) and Smets and Wouters (1999). By focusing on interest rates for the identification of monetary policy shocks, we deviate from the structural VAR literature cited in the previous paragraph, which attaches the role of the monetary policy instrument to money supply. While this is a point in common with Mojon and Peersman (2003), our use of non-recursive identification restrictions marks a difference with the latter two authors who instead employ a Cholesky identification approach. 14 Table 6 reports the accumulated responses from oil shocks derived from this alternative approach (panel A) alongside counterfactual experiments (panel B). With regard to the former, we observe that impulse response results are broadly in line with those resulting from the Cholesky ordering as shown above in Table 4. In particular, the short-term interest rate responses to an exogenous oil price hike again suggest that monetary policy has responded by tightening since 1986. 15 The impulse response for the nominal interest rate points to an only very gradual increase that is perceived only between the first two years and the third. The degree of smoothness implied by our results are even stronger than the already very high one detected by López s (2002) Taylor rule estimation over 1984-1998. In contrast to the period since 1986, it appears that monetary policy has been accommodative in the full sample. Based on our short-run restriction identification, we perform a monetary policy counterfactual exercise. The latter assumes that the real short-term interest rate does 13 We thus assume that policymakers have serious difficulties assessing ongoing real-side developments within the quarter. This is a reasonable assumption given the lack of reliable indicators for both GDP components and sectoral output beyond the relatively small industrial sector. 14 We also differ from Mojon and Peersman (2003) in that they need to include German interest rate shocks to eliminate the price puzzle, which arises when monetary policy tightening say induces higher inflation. Instead, this puzzle is absent from our results regardless of whether such German shocks are included or not. This is in line with previous studies including commodity prices (such as oil prices here) to get rid of this puzzle. Another puzzle usually featuring in structural VARs is the exchange rate puzzle (interest rate hikes - say - lead to an exchange rate depreciation). While this puzzle is present in our full sample results, the evidence for post-1986 sample is mixed, with the puzzle showing only for some time horizons. All of these results are available from the authors upon request. 15 This result can be interpreted as a more price-stability-oriented monetary policy approach in Spain following EU membership. One factor playing a role at an early stage in the post-1986 period is the monetary tightening 1989-1992 phase over which the antiinflationary bias inflation was reinforced via REER appreciation of the peseta (see Camarero and Tamarit, 2002). 9

not react to the shock. The counterfactual results, which are also shown in panel B of Table 6, are reasonable. They indicate that, in the period since 1986, where the prediction is that of tightening in monetary policy, the counterfactual projection indicates higher values of both real output growth and inflation compared to the baseline scenario. In quantitative terms, our counterfactual results suggest that the fraction of macroeconomic variability induced by the oil shock indirectly via monetary policy changes appears to be rather modest, in line with Hamilton and Herrera s (2004) results for the US. In contrast to our results for the period since 1986, in the full sample case the counterfactual exercise points to an accommodative policy, as the implied projection consists of lower levels of both real output growth and inflation. 4. Concluding remarks This paper studies the role of oil price shocks in shaping interest rates in Spain. The main empirical findings for the period since Spain s EU accession in 1986 are the following. The preferred specification is the linear model, which is found to beat leading non-linear approaches. Real short-term interest rates tend to rise following an unexpected oil price hike, while nominal interest rates increase very gradually. The share of real interest rate variability explained by oil prices is around 10%. Historical decompositions show that oil prices impacted both real and nominal interest rates at around the 1990 oil price spike. With regard to the latest episode of high oil prices, nominal interest rates are found to be pushed upwards at two times, namely, in 1999 and 2005-2006. In light of higher inflationary pressures, the increase of nominal interest rates in 1999 appears not to have prevented a fall in real rates, which are found to rise in response to the shock only in 2000-2001. Turning to counterfactual analysis, the period since 1986 can be characterised as one of tightening of interest rates in response to oil shocks, with interest rate hikes contributing to reducing inflation and making the economic slowdown more marked. It is worth saying, however, that the fraction of macroeconomic variability induced by the oil shock indirectly via monetary policy changes appears to be rather modest. The main results from this paper are not robust to estimating the model over a longer sample going back to 1970. Over this sample, it is found that oil prices have nonlinear impacts, real interest rates are little affected by oil shocks and monetary policy appears to be accommodative. With regard to the latter, the evidence conclusively shows that interest rates were cut at the time of the sizeable oil price increases of 1974-1975. Acknowledgements The views expressed in this paper are those of the authors and do not necessarily reflect the position of the European Central Bank. The usual disclaimer applies. 10

References Alonso, F., Ayuso, J., Martínez-Pagés, J. The informational content of inflation for Spanish interest rates. Investigaciones Económicas 2000;24; 455-471. Ayuso, J., Escrivá, J. Trends in the monetary policy strategy in Spain. In: Malo de Molina, J. Viñals, J., Gutiérrez, F. (Eds.), Monetary policy and inflation in Spain. New York; St. Martin s Press; 1998. p. 131-158. Barrell, R., Pomerantz, O. Oil prices and the world economy. NIESR Discussion Paper No. 242; 2004. Bernanke, B., Gertler, M., Watson, M. Systematic monetary policy and the effects of oil price shocks. Brookings Papers on Economic Activity 1997;1; 91-158. Camarero, M., Ordóñez, J., Tamarit, J. Monetary transmission in Spain: a structural cointegrated VAR approach. Applied Economics 2002;34; 2201-2212. Camarero, M., Tamarit, J. A panel cointegration approach to the estimation of the peseta real exchange rate. Journal of Macroeconomics 2002;24; 371-393. Cuñado, J., Pérez de Gracia, F. Do oil price shocks matter? Evidence from some European countries. Energy Economics 2003;25; 137-154. Cushman, D., Zha, T. Identifying monetary policy in a small open economy under flexible exchange rates. Journal of Monetary Economics 1997;39; 433 448. De Miguel, C., Manzano, B., Martín-Moreno, J. Oil price shocks and aggregate fluctuations. The Energy Journal 2003;24; 47-61. De Miguel, C., Manzano, B., Martín-Moreno, J. Oil shocks and the business cycle in Europe. In: De Miguel, C., Labandeira, X., Manzano, B. (Eds.), Economic modelling of climate change and energy policies. Edward Elgar Publishing; 2006. Elliott, G. Efficient tests for a unit root when the initial observation is drawn from its unconditional distribution. International Economic Review 1999;40; 767--83. Elliott, G., Rothenberg, T., Stock, J. Efficient tests for an autoregressive unit root. Econometrica 1996;64; 813-36. Hamilton, J. This is what happened to the oil price-macroeconomy relationship. Journal of Monetary Economics 1996;38; 215-220. Hamilton, J. Oil and the macroeconomy. In: Durlauf S., Blume, L. (Eds.), New Palgrave Dictionary of Economics and the Law, 2nd ed. London: Macmillan; 2006 (forthcoming). Hamilton, J., Herrera, A. Oil shocks and aggregate macroeconomic behavior: The role of monetary policy. Journal of Money, Credit and Banking 2004;36; 265-286. Hooker, M. What happened to the oil price-macroeconomy relationship? Journal of Monetary Economics 1996;38; 195-213. Jiménez-Rodríguez, R., Sánchez, M. Oil price shocks and real GDP growth: Empirical evidence for some OECD countries. Applied Economics 2005;37; 201-228. Kim, S., Roubini, N. Exchange rate anomalies in the industrial countries: A solution with a structural VAR approach. Journal of Monetary Economics 2000;45; 561-586. Lee, K., Ni, S., Ratti, R. Oil shocks and the macroeconomy: The role of price variability. The Energy Journal 1995;16; 39-56. López, V. Has Banco de España followed a Taylor rule using real-time information? Investigaciones económicas 2002;26; 475-496. Mojon, B., Peersman, G. A VAR description of the effects of monetary policy in the individual countries of the euro area. In: Angeloni, I., Kashyap, A., Mojon, B. (Eds.), Monetary transmission in the euro area. Cambridge: Cambridge University Press; 2003. p. 56-74. 11

Mork, K. Oil and macroeconomy when prices go up and down: An extension of Hamilton's results. Journal of Political Economy 1989;97; 740-744. Mork, K., Olsen, O., Mysen, H. Macroeconomic responses to oil price increases and decreases in seven OECD countries. The Energy Journal 1994;15; 19-35. Smets, F. Measuring monetary policy shocks in France, Germany and Italy: The role of exchange rate. Swiss Journal of Economics and Statistics 1997;133; 597-616. Smets, F., Wouters, R. The exchange rate and the monetary transmission mechanism in Germany. The Economist 1999;147;. 489-521. 12

Tables 1 to 6 Table 1: Unit-root tests Model with constant and trend Model with constant Model without constant ADF DFGLS P T DFGLS U Q T ADF DFGLS P T DFGLS U Q T ADF DFGLS Real GDP Levels -2.98-2.38 7.23-2.79 2.17 0.08 1.85 222.5 0.42 41.33 2.76 2.76 First Log-diff. -3.37* -3.38** 5.03** -3.53** 1.58*** -3.46** -2.46** 3.54** -3.50*** 1.51*** -2.08** -2.08** REER Levels -3.08-2.08 8.05-2.66 2.88* -2.88* -0.49 16.86-2.69* 6.74 0.93 0.93 First Log-diff. -4.80*** -4.33*** 0.77*** -4.90*** 0.18*** -4.78*** -3.87*** 0.53*** -4.79*** 0.06*** -4.68*** -4.68*** CPI Levels -2.25-1.74 1.88*** -2.26 2.12** -3.86*** -0.47 6.34-1.58 18.77 0.19 0.19 First Log-diff. -2.97-1.08 36.74-1.77 11.41-0.81-0.89 11.90-0.85 22.15-0.75-0.75 Real wage Levels -2.23 0.39 247.7-0.59 80.64-5.13*** 2.13 652.5-2.66* 299.3-6.82*** -6.82*** First Log-diff. -14.26*** -14.26*** 1.53*** -14.03*** 0.83*** -1.78-0.88 34.95-1.83 45.50-1.77* -1.77* Real short-term Levels -1.30-1.47 20.28-1.47 10.48-1.47-1.44 5.90-1.48 10.69-1.40-1.40 interest rate First diff. -6.73*** -2.40 24.53-4.49*** 2.14** -6.71*** -1.14 16.27-5.68*** 0.01*** -6.74*** -6.74*** Real long-term Levels -1.29-1.28 22.94-1.34 11.57-1.28-1.26 6.60-1.29 12.51-1.22-1.22 interest rate First diff. -5.67*** -2.26 40.73-3.99*** 1.96*** -5.68*** -1.03 25.43-4.78*** 0.16*** -5.70*** -5.70*** Real oil price Levels -2.02-1.21 29.10-1.58 11.29-2.03-0.02 29.23-1.91 12.86 0.74 0.74 First Log-diff. -6.50*** -6.45*** 0.06*** -6.43*** 0.03*** -6.53*** -6.51*** 0.01*** -6.52*** 0.02*** -6.46*** -6.46*** The sample is 1970:1-2006:1 for the variables in levels, and starts one quarter later for the variables in first differences. We use data-driven lag selection procedures for the Augmented Dickey-Fuller tests, taking 1.645 as the critical value used for significance of lagged terms and 8 as the maximum number of lags allowed in these procedures into account. The same number of lags is used in the other tests considered. We denote with one/two/three asterisks the rejection of the null hypothesis at a 10%/5%/1% critical level. Critical levels used for ADF test are the following: - In the model with constant and trend: -4.05 (1%), -3.45 (5%) and -3.15 (10%). - In the model with constant: -3.50 (1%), -2.89 (5%) and 2.58 (10%). - In the model without constant: -2.59 (1%), -1.94 (5%) and 1.62 (10%). Critical levels used for DFGLS test are the following: - In the model with constant and trend: -3.48 (1%), -2.89 (5%) and 2.57 (10%). - In the model with constant: -2.58 (1%), -1.95 (5%) and 1.62 (10%). - In the model without constant: -2.58 (1%), -1.95 (5%) and 1.62 (10%). Critical levels used for PT test are the following: - In the model with constant and trend: 3.96 (1%), 5.62 (5%) and 6.89 (10%). - In the model with constant: 1.99 (1%), 3.26 (5%) and 4.48 (10%). Critical levels used for DFGLSU test are the following: - In the model with constant and trend: -3.71 (1%), -3.17 (5%) and 2.91 (10%). - In the model with constant: -3.28 (1%), -2.73 (5%) and 2.46 (10%). Critical levels used for QT test are the following: - In the model with constant and trend: 2.05 (1%), 2.85 (5%) and 3.44 (10%). - In the model with constant: 3.06 (1%), 4.65 (5%) and 5.94 (10%).

Table 2: Likelihood ratio tests LINEAR ASYMMETRIC SCALED OIL PRICE NET OIL PRICE o t o t + o t - SOPI t SOPD t NOPI t NOPD t 1986:1-2006:1 2.3E-008*** 3.0E-016*** 1.3E-015*** 2.0E-014*** 3.2E-016*** 1.7E-011*** 1.1E-016*** 1970:2-2006:1 2.2E-005*** 2.0E-08*** 0.12225 1.5E-007*** 0.19521 4.4E-007*** 0.02136** 1970:2-1998:4 1.2E-007*** 0.00028*** 0.47977 0.00222*** 0.65855 0.01061** 0.00672*** One/two/three asterisks mean a p-value less than 10%/5%/1%.

Table 3: Relative performance of the models Linear Asymmetric Scaled Net 1986:1-2006:1 AIC 13.133 18.922 17.240 16.649 BIC 22.031 30.509 28.828 28.237 1970:2-2006:1 AIC 18.846 18.301 18.042 24.797 BIC 25.231 24.686 24.749 33.112 1970:2-1998:4 AIC 19.411 19.003 18.126 28.384 BIC 26.843 23.958 23.081 34.828 The results for the periods 1970:2-2006:1 and 1970:2-1998:4 are based on a seven-variable VAR that excludes oil price decrease variables from all three non-linear models, namely, the asymmetric, scaled and net specifications, with one exception: the Spain s net model, where oil price decrease variable is found to be statistically significant for the system. For the period 1986:1-2006:1, the results are based on an eight-variable VAR that includes oil price decrease variables from all three non-linear models, namely, the asymmetric, scaled and net specifications.

Table 4: Accumulated Responses in the baseline VAR model 1986:1-2006:1 1970:2-2006:1 1970:2-1998:4 Variables after 1 year after 2 years after 3 years after 1 year after 2 years after 3 years after 1 year after 2 years after 3 years Real GDP growth -1.5-4.2-5.1-2.9-7.3-6.4-1.9-5.9-6.8 Inflation 1.0 1.9 1.1 1.5 0.9 2.2 0.3 2.8 5.8 Real short-term interest rate 0.7-0.2 0.9-1.1-3.1-7.5 0.0-4.0-7.6 Nominal shortterm interest rate 1.7 1.7 2.0 0.4-2.2-5.3 0.3-1.2-1.8 The entries refer to the accumulated responses attributed to a 100% oil price shock. The models used are: linear model for the period 1986:1-2006:1and scaled models for 1970:2-2006:1 and 1970:2-1998:4. Identification is achieved by ordering the oil price variable first in the VAR system.

Table 5: Forecast error variance decomposition 1986:1-2006:1 1970:2-2006:1 1970:2-1998:4 Variables after 1 year after 2 years after 3 years after 1 year after 2 years after 3 years after 1 year after 2 years after 3 years Real GDP growth 6.9 15.6 16.3 8.0 16.2 15.9 6.6 14.2 13.9 Inflation 9.1 10.2 10.8 4.4 5.2 5.2 2.0 2.5 2.6 Real short-term interest rate 10.5 9.6 9.4 3.0 3.6 5.8 1.8 3.1 3.8 The entries refer to the fraction of each variable's variance attributed to oil price shocks (in percentage). The models used are: linear model for the period 1986:1-2006:1and scaled models for 1970:2-2006:1 and 1970:2-1998:4. Identification is achieved by ordering the oil price variable first in the VAR system.

Table 6: Accumulated Responses using short-run restrictions and counterfactuals 1986:1-2006:1 1970:2-2006:1 Variables after 1 year after 2 years after 3 years after 1 year after 2 years after 3 years A) Shor--run restrictions Real GDP growth -1.6-5.1-6.8-3.0-8.5-9.6 Inflation 1.1 2.4 1.9 1.6 1.3 2.9 Real short-term interest rate 0.7-0.1 1.1-1.1-3.4-9.0 Nominal short-term interest rate 1.8 2.3 3.0 0.5-2.1-6.1 B) Counterfactuals (unchanged real short-term interest rate) Real GDP growth -1.3-4.6-6.4-2.9-9.2-10.9 Inflation 1.6 2.9 2.2 1.5 1.0 0.8 Real short-term interest rate 0.0 0.0 0.0 0.0 0.0 0.0 Nominal short-term interest rate 1.6 2.9 2.2 1.5 1.0 0.8 The entries refer to the accumulated responses attributed to a 100% oil price shock. The models used are: linear model for the period 1986:1-2006:1and scaled model for 1970:2-2006:1. The near-var system used here does not allow for feedback to oil prices from lags of the remaining variables. Identification is achieved by means of short-run restrictions in the VAR system. For details, see the main text.

Figures 1 and 2 Figure 1. Historical decompositions of real short-term interest rates Estimation over 1986:1-2006:1 0.8 0.6 0.4 0.2 0.0-0.2-0.4-0.6-0.8-1.0 1989 1991 1994 1996 1999 2001 2004 3.0 Estimation over 1970:2-2006:1 2.0 1.0 0.0-1.0-2.0-3.0 1973 1975 1978 1980 1983 1985 1988 1990 1993 1995 1998 2000 2003 2005

Figure 2. Historical decompositions of nominal short-term interest rates Estimation over 1986:1-2006:1 0.8 0.6 0.4 0.2 0.0-0.2-0.4-0.6 1989 1991 1994 1996 1999 2001 2004 2.0 Estimation over 1970:2-2006:1 1.5 1.0 0.5 0.0-0.5-1.0-1.5 1973 1975 1978 1980 1983 1985 1988 1990 1993 1995 1998 2000 2003 2005