Global Oil Prices and the Macroeconomy: the Role of Tradeable Manufacturing versus Nontradeable Services Makram Khalil∗ Deutsche Bundesbank

August 28, 2017

Abstract: This paper identifies shocks to the demand for manufactured goods (tradeables) as a key driver of global oil prices. In an estimated two-country model (US and the rest of the world) with a tradeable oil-intensive sector (manufacturing) and a nontradeable sector (services) in each country, manufacturing demand shocks can explain comovement between oil prices and the cyclical wedge between manufacturing and services output as well as aggregate inflation and policy rates in the US. These dynamics are found to rationalize the observed pattern during important historical episodes in the relationship between oil prices and the US macroeconomy. Keywords: Endogenous Global Oil Price, Trade Channel, Manufacturing and Services, Oil and the Business Cycle, Open Economy JELs: E32, F41, Q43

I would like to thank participants in the European Summer Meeting of the Econometric Society 2016, the 2nd Annual Conference of the International Association for Applied Econometrics 2015, the Computing in Economics and Finance 2015 conference, the 7th FIW Conference on International Economics, as well as seminar participants at the University of Vienna for useful discussions and valuable comments. Supported by funds of the Oesterreichische Nationalbank (Oesterreichische Nationalbank, Anniversary Fund, project number: 15979). Contact: Deutsche Bundesbank, Wilhelm-Epstein-Straße 14, 60431 Frankfurt am Main, E-Mail: [email protected]. This paper represents the author’s personal opinions and does not necessarily reflect the views of the Deutsche Bundesbank or its staff. ∗

1

1

Introduction

Large up and downswings in global oil prices since the mid-2000s have renewed the interest in causes and consequences of such movements. There is growing consensus in the literature about the need to differentiate between several underlying sources of oil price movements when analyzing the effects on macroeconomic outcomes (cf. inter alia Kilian 2009, Bodenstein et al. 2012). The trade channel plays thereby an important role, especially in case of disturbances to the oil price that stem from a rise in economic activity. When, for example, Asian industrialization is contributing to a boom in the global business cycle, then this fuels global oil prices but at the same time has positive effects on US economic activity because it leads to rising non-oil exports. The recent literature considers models where there is only one tradeable sector, a practice that has to be questioned when it comes to analyze the relationship between the global oil price and the macroeconomy. In a period in which, for example, Asian growth fuels global oil prices, the US manufacturing sector is still positively affected from increasing foreign demand because of more exports. The large and important US services sector – that is barely exporting – does, however, not benefit from increasing foreign demand. Additionally, during such episodes this sector has to face inflationary pressure triggered by the manufacturing sector. If, furthermore, one acknowledges that the global oil price is driven by global demand, then this should be attributed mainly to the global production of tradeable manufacturing goods. Manufacturing is a sector that is, compared to services, very oil-intensive in production; also transporting manufactured goods heavily relies on cargo vessels and trucks that typically use oil as input. In a nutshell, I argue in this paper that global manufacturing production can be considered as the key demand force of changing global oil prices and importantly, the oil / macroeconomy relationship depends crucially on how the large domestic nontradeable services sector is affected by shocks that drive manufacturing production. I build a structural two-country model (US and the rest of the world) with a global oil market and two broad production sectors in each economy facing nominal rigidity (manufacturing and services). In the model, manufacturing is the more oil-intensive 2

sector and produces traded goods, whereas services is the nontradable sector.1 The model incorporates a rich shock structure and is estimated with global oil market, global industrial activity and US sectoral data (1974 until 2007). This allows to uncover changes in the wedge between production in manufacturing and services as an important dimension of macroeconomic effects of oil-related shocks that has not been considered so far. The main result of the paper is that the global oil price is quantitatively importantly related to cyclical divergences between manufacturing and services. Conducting historical decomposition, I identify (home and foreign) manufacturing demand shocks as the key demand-type driver of global oil prices. These shocks can explain the joint occurrence of a rising global oil price, a wedge between expanding manufacturing and contracting services output in the US, rising inflation in both of these sectors and rising US nominal interest rates. Interestingly, such a cyclical pattern is found during many episodes of rising oil prices throughout the sample, for example, around the Volcker recessions in the early 80’s, the episode around the 2001 recession or during 2003-2008. The channel can also rationalize the cyclical pattern observed in times of falling oil prices like for example in 1986. Positive manufacturing demand shocks have, moreover, interesting implications for aggregate macroeconomic outcomes. In case of foreign manufacturing demand shocks US aggregate output is found to barely respond, and for domestic manufacturing demand shocks there are rather small positive output effects. In both cases aggregate inflation rises because of inflationary impacts originating in the manufacturing sector. The oil-intensive manufacturing sector is supporting aggregate output in case of manufacturing demand shocks despite the rise of oil prices. More demand for tradeable manufacturing goods leads to higher output in this sector at home and abroad and a reallocation of labour from the services sector to the manufacturing sector. The results challenge the popular view that output of oil-intensive industries contracts first when oil prices rise. The described mechanism is found to affect the US economy similarly in the 70’s compared 1

Cf. Table 5 in the appendix for descriptive statistics on oil-use and tradeability of the US manufacturing vs. the US services sector. The data shows that the US manufacturing sector is intensively exporting and using oil, while the US services sector does barely export and is not very oil-intensive.

3

to the 2000’s. The findings add to the literature that describes shocks to oil-efficiency (respectively oil-intensity) as the most important demand-type source of global oil price movements (cf. Bodenstein et al. 2011 or Bodenstein et al. 2012; also discussions in Kilian and Hicks 2013). Distinguishing between manufacturing and services allows to directly identify expansions of oilintensive manufacturing production as a main driver of global oil prices, which provides a more elaborated rationale than the one discussed in the literature. Previous studies often assume that an increase of the global oil price driven by global activity is – from the point of view of the US – always like an exogenous oil supply shock (cf. inter alia Blanchard and Gali 2010 and Blanchard and Riggi 2013). In this view the trade channel does not play much of a role and the direct effects of changing oil prices dominate, which is challenged by the results presented here. In the estimated model, negative oil supply shocks are contractionary for both sectors and thereby the US aggregate economy, while inflation goes up. For positive manufacturing demand shocks, inflation picks up much stronger relative to the oil price rise, furthermore output is not contracting. The effects of rising oil prices do clearly not dominate. Still, because of decreasing services production, the positive response of aggregate output is relatively modest if not negligible. The resulting historical shock structure of the estimated model does well relate to corresponding shocks obtained by the parsimonious SVAR model employed by Kilian (2009), which provides an important robustness check. Also, the model has good out-of-sample explanatory power in matching the real oil price path between 2008 and 2016. In doing so, it outperforms a similar model that has only one (tradeable) production sector. The paper contributes to a large number of studies that analyze the relationship between the oil price and the macroeconomy. Most prominently, the empirical literature establishes the stylized fact that US recessions are typically preceded by rising oil prices (cf. Hamilton, 1983, Hamilton, 2011). Many contributions incorporate oil prices in DSGE models in order to study the effects of exogenous oil price shocks in a closed economy setting (inter alia Leduc and Sill, 2004; Carlstrom and Fuerst, 2006; Natal, 2012). 4

The role of openness in the context of macroeconomic effects of oil prices was first addressed theoretically by Backus and Crucini (2000) in a flexible price model and later in models with sticky prices (cf. Balke et al. 2010, Bodenstein et al. 2011 and Bodenstein et al. 2012). None of these papers includes a nontradeable sector in the analysis. The remainder of this study is organised as follows. The detailed model structure is described in section 2, while the data and structural estimation approach are discussed in section 3. Section 4 focuses on details of the most relevant transmission mechanisms by employing impulse response analysis. Section 5 presents shock decomposition for important historical episodes. In section 6 forecasting exercises for the episode after 2008 and until 2016 are conducted. Finally, section 7 concludes.

2

Manufacturing and Services in a Model of Endogenous Global Oil Prices

In the model there are two countries with each country having a tradeable (manufacturing) and a nontradeable (services) sector (in the spirit of Stockman and Tesar 1995). The two countries (Home and Foreign) are of different size. I build on the New Open Economy Macroeconomy (NOEM) literature that introduces sticky prices in such a framework (cf. inter alia Rabanal and Tuesta, 2013). Agents around the world demand services and manufacturing. The domestic services demand is fully covered by domestic service producers but there is a preference for home and foreign produced manufactured goods, i.e. manufactured goods are traded across countries. Each of the two countries imports oil from a global oil supplier and the global oil price is determined endogenously by global factor demand (i.e. home and foreign manufacturing and services production) as well as exogenous supply. Profits of the oil supplier are distributed proportionally to the two countries as lump sum transfers (similar to the specification of Campolmi, 2008). The interesting shocks in the model are demand-type shocks (especially demand shocks to foreign or domestic manufacturing output) as an alternative to supply shocks (foreign and domestic technology shocks) and exogenous shocks to the global oil production. All the shocks 5

have in common that they alter global oil prices, however, the channels of transmission differ substantially. Decision making of households and firms is relatively standard. Modeling global oil markets puts another building block to the model. The home and the foreign economy have the same structure. In the following the home economy is described and, if not indicated differently, the same relations hold equivalently for the foreign economy. Foreign variables are denoted by an asterisk.

2.1

The Model

Household’s Consumption The domestic representative agent consumes each period the following consumption goods bundle h

Ct ≡ (1 − γ)

1/φ

(CT,t )

φ−1 φ

+γ

1/φ

(CN,t )

φ−1 φ

φ i φ−1

(1)

where γ is the share of nontradeable goods in the consumption basket and φ is the elasticity of substitution between final manufactured goods (tradeables) CT,t and final services (nontradeables) CN,t . The corresponding consumer price index is given by  1  Pt = (1 − γ)(PT,t )1−φ + γ(PN,t )1−φ 1−φ . The utility function of the representative agent is separable in consumption Ct and labour Nt ∞

U0 = E 0

X t=0

N 1+ϕ β ψt ln Ct − t 1+ϕ t





(2)

where ϕ is the inverse elasticity of labour supply with respect to the real wage and 0 < β < 1 denotes the constant discount factor. ψt is a preference shock that follows an AR(1) process in logs.

6

International Financial Market and Household’s Budget Constraint Financial trade is introduced in a standard way (cf. inter alia Benigno and Thoenissen 2003) by the assumption that a domestic household can hold domestic and foreign bonds, whereas a foreign household can only hold foreign bonds. The net foreign bond holdings of domestic households determine the trade balance of the two countries. Assuming full use of resources, the inter-temporal budget constraint of the domestic household is given by St BF,t BH,t−1 St BF,t−1 Wt BH,t + + + Nt −Ct +ΠT H,t +ΠN,t +Dt +Tt , = St ,BF,t ∗ Pt Rt Pt Rt Φ( Pt Pt Pt ) Pt Y t (3) where BH,t and BF,t denominate holdings of home and foreign bonds, Rt and Rt∗ are home and foreign gross nominal interest rates, respectively, the function Φ gives a small financial intermediary cost2 , St is the nominal exchange rate (units of home currency in units of foreign currency), Wt denotes the nominal wage, Dt denote lump-sum transfers from the global oil supplier (cf. below), ΠT H , ΠN are profits from the tradeable and the nontradeable sector and Tt denote lump sum transfers, respectively. Foreign agents gain dividends from the financial intermediary and do only have access to foreign bonds, so the budget constraint of a foreign agent differs with that respect. Final Goods Consumption Bundling It is assumed that at the border there is a consumption bundler that aggregates home and foreign produced tradeables according to following technology i θ h θ−1 θ−1 θ−1 C˜T,t ≡ (1 − δ)1/θ (CT H,t ) θ + δ 1/θ (CT F,t ) θ 2

(4)

The introduction of this cost ensures stationarity (cf. Schmitt-Grohe and Uribe 2003). Like in Benigno (2009), I assume that the cost function Φ takes the value 1 when the net foreign asset position approaches its steady-state value which is assumed to be zero. The function is differentiable and decreasing in the neighborhood of zero.

7

where δ denotes share of foreign goods in the consumption of manufacturing and θ is the elasticity of substitution between home CT H,t and foreign manufactured goods CT F,t . The two countries are of unequal size with the population size in Home given by n and and in Foreign by 1 − n. The varieties of home tradeable goods are indexed by fT H ∈ [0, n), while the varieties of foreign tradeable goods are indexed by fT F ∈ [n, 1]. The respective varieties are aggregate with a standard CES technology. Firms Domestic production takes place with a CES technology such that output in a single firm fl is produced according to i τ h τ −1 τ −1 τ −1 Yl,t (fl ) = Zl,t (1 − αl )Nl,t (fl ) τ + αl Ol,t (fl ) τ

(5)

for l ∈ {T H, N } denoting a sector, fl ∈ [0, n) is a domestic firm producing a specific variety, αl is the quasi share parameter for oil use (same for all firms within a sector), Zl,t denotes sector specific technology and Nl,t (fl ) and Ol,t (fl ) denote firm specific labour and oil demand. The parameter τ gives the elasticity of substitution between the two factor inputs. The Hicks-neutral sector specific technology Zl,t evolves according to an AR(1) process in logs. Price setting is introduced a la Calvo, i.e. in each period only a fraction of (1 − ϕl ) ∈ [0, 1] firms is allowed to readjust their prices. It is assumed that the law of one price holds for internationally traded goods. Trade Balance and Real Exchange Rate The trade balance evolves according to

St BtF Pt Rt∗ Φ(

St BtF Pt

)

=

F St Bt−1 +N Xt , Pt

N Xt =

CT∗ H,t − PT F,t CT F,t PT H,t 1−n n (6) Pt

where N Xt denotes the real value of net exports. 8

The real exchange rate and the terms of trade are defined as

Qt ≡

St Pt∗ , Pt

Tt ≡

PT F,t . St PT∗H,t

(7)

The Global Oil Market There is a global oil supplier that belongs to both countries proportionally so that the same per-capita profits are redistributed across countries (cf. Campolmi, 2008). Oil is assumed to be non storeable and the global oil supply Osupply,t follows an AR(1) process in logs. The global oil demand is determined in each period by factor demand of the tradeable (manufacturing) and the nontradeable (services) sector of each of the two countries. It is assumed that the law of one price holds for oil prices and that the elasticity of substitution between oil and labour is the Po same in both countries. Then in equilibrium the real oil price Ptt is

 τ  τ Wt Wt αN αT H = NT H,t + n NN,t + × [n Osupply,t 1 − α T H Pt 1 − α N Pt τ τ   ∗ αN αT∗ H Qt Wt∗ Qt Wt∗ ∗ ∗ NT F,t + (1 − n) NN,t ]. + (1 − n) ∗ ∗ ∗ ∗ 1 − αT H Pt 1 − αN Pt (8)



Pto Pt

τ

1

Monetary Policy Monetary policy follows a Taylor-type rule, i.e. the central bank targets consumer price inflation and the output growth rate

Rt = R

(1−νr )

νr Rt−1



Pt /Pt−1 Π

(1−νr )κπ

(Yt /Yt−1 )(1−νr )κy exp (ξr )

(9)

where νr characterizes interest rate smoothing, ξr is a monetary policy shock and κπ and κy denote the Taylor rule coefficients for inflation and output growth, respectively.

9

Market Clearing The market for services goods clears such that YN,t = CN,t + GN,t ,

GN,t+1 = (GN,t )ρGN exp(ξGN ,t+1 )

∗ ∗ YN,t = CN,t + G∗N,t ,

G∗N,t+1 = (G∗N,t )

ρG ∗

N

exp(ξG∗N ,t+1 )

(10)

(11)

where GN and G∗N are sector-specific demand shocks to services demand. The market for manufactured goods clears such that YT,t = CT H,t +

1−n ∗ CT H,t , n

∗ YT,t = CT∗ F,t +

n CT F,t . 1−n

(12)

Total manufacturing demand evolves in the following way C˜T,t = CT,t + GT,t ,

GT,t+1 = (GT,t )ρGT exp(ξGT ,t+1 ).

∗ ∗ C˜T,t = CT,t + G∗T,t ,

G∗T,t+1 = (G∗T,t )

ρG ∗

T

exp(ξG∗T ,t+1 ).

(13)

(14)

where GT and G∗T are shocks to manufacturing demand. The interpretation of this sector specific shock can be a government-led increase in manufacturing output (financed by lump-sum taxation). More generally, it explains an unexpected shifts towards demand for manufactured goods that can not be explained through changes in the interest rate or other relative prices. The model is closed by defining aggregate real domestic output (real GDP) as the sum of the value of the sectoral outputs PY t PT H YT,t + PN,t YN,t Yt = . Pt Pt

(15)

where PYt is the nominal domestic producer price index defined as an output share weighted sum of the sectoral nominal prices.3 3

In general, PYt is not necessarily equal to the CPI Pt . In steady state the two coincide since the trade balance is zero. Cf. Ferrero et al. 2010.

10

2.2

Solution Approach

The solution approach is to log-linearize the model around a deterministic steady state and to solve the linear model using perturbation methods. In steady state, international bond holdings are zero, international trade is balanced and inflation is zero. All price resetting firms set the same price and production is subsidized by fiscal transfers such that the distortion from monopolistic competition is offset.

3

Data and Structural Estimation

The model is estimated structurally employing a Bayesian approach and with data of the US economy, the global oil market as well as global industrial activity.4

3.1

Data

US data for inflation and output in the manufacturing and services sector as well as interest rates are available for a long time span including important episodes in the global oil market. There is, however, a lack of data for the same variables of the foreign block. This problem concerns for example data on the Chinese economy but in general disaggregated data is not available for the whole time span. In order to identify global manufacturing output I use a global industrial activity index as a proxy. The literature on endogenous oil price movements uses various measures to capture the global business cycle. Kilian (2009) introduces an index of real global activity which is based on shipping rates for major commodities. Another possible measure is the global production index employed by Beidas-Strom and Pescatori (2014). I employ the latter one because like the measure of US manufacturing output it is a quantity based index, while the Kilian 4

For a detailed explanation of the estimation approach the reader is referred to An and Schorfheide (2007) among many other sources. An early example for an application of a Bayesian estimation of a two-country model is Lubik and Schorfheide (2005). Other contributions include inter alia Rabanal and Tuesta (2010) and Rabanal and Tuesta (2013). The latter study incorporates a nontradeable sector.

11

index is based on prices.5 It is assumed that the index is proportional to global industrial activity and to global trade and corresponds to the production of global tradeables in the DSGE model. The sample is restricted by the availability of the refiner import oil price series from 1974:Q1. The estimation of the model presented in the following is conducted for the time span 1974:Q1 to 2007:Q4. The forecasting exercise in section 6 extends the analysis to the period of 2008:Q1 until 2016:Q4. The data sources and the observable variables are reported in the appendix. The observable variables have to be mapped into the stochastic stationary model which is log-linearized around a deterministic steady state and variables have the interpretation of percentage deviations from steady state. Following Smets and Wouters (2003), the non-stationary series (real output in home services and manufacturing, global industrial production as well as global oil production) are transformed by linearly detrending the logarithm of the observed time series. The inflation variables, the real oil price (in log level) as well as the domestic interest rate are assumed to be stationary, so in these cases the original data is not detrended. All variables are demeaned in order to match the steady state values of the model. The model counterpart of observed global industrial production index is the country size weighted sum of domestic and foreign manufacturing real output.

3.2

Calibration

As common in models of that type some parameters remain calibrated. The discount factor is set to β = β ∗ = 0.99 and the inverse of the Frisch elasticity of labour to ϕ = ϕ∗ = 2 which is within the range of estimates from microdata. The size of the home economy (US) is calibrated to n = 0.29 which refers to the share of US GDP in global GDP.6 The steady state weight of 5

As in Beidas-Strom and Pescatori (2014) the series for the global activity index is constructed by splicing the global industrial production index provided by the Netherlands Bureau for Economic Policy Analysis (CPB) – which is available only from 1991:Q1 onward – to the index of industrial production in OECD countries from 1974:Q1 to 1990:Q4. For the year 1974 I use vintage data of the Dallas Fed in order overcome data limitations for the original OECD index. 6 The share of US GDP in global GDP is measured using data from IMF world economic outlook for the time span 1980-2007. Nominal GDP in all countries is measured

12

services in consumption expenditure is calibrated to γ = γ ∗ = 0.60.7 The steady state share of foreign manufacturing in total manufacturing expenditure is set to δ = 0.38 in order to match the observed share of exports in the manufacturing sector (cf. Table 5 in the appendix). In order to be consistent with balanced trade in steady state and relative prices of one the share of imported intermediate goods is calibrated to δ ∗ = nδ/(1 − n) = 0.16. Consistent with the estimates of Rabanal and Tuesta (2010) the financial intermediation cost is set to χ = 0.007. The oil cost share is set to ∗ α ˜T = α ˜ T∗ = 0.060 in manufacturing and α ˜N = α ˜N = 0.015 in services which corresponds to values obtained from US input-output tables (cf. Table 5 in the appendix). The relative importance of the sector-specific demand G∗ shock is ιT = GYTT = Y T∗ = 0.1.8 The elasticity between oil input an labour T in the production function is set to τ = τ ∗ = 0.09. The low elasticity – which is within the range of estimated values reported by Hamilton (2009) – reflects that oil can not be easily substituted in the short run.9 All remaining parameters and the autocorrelation and standard deviations of the foreign as well as domestic shock processes are estimated (cf. below).

3.3

Bayesian Prior and Posterior Estimation Results

Altogether 25 parameters are estimated. It is assumed that the elasticity between services (nontradeables) and manufacturing (tradeables), the elasin US-Dollar. 7 It should be noted that the implied share of services in total output is lower than indicated by recent input-output tables (cf. Table 5 in the appendix). The calibration approach excludes wholesale and retail trade from services (cf. Dotsey and Duarte, 2008), furthermore the average share over the whole sample is lower than more recent values because of a well known structural change in the US economy. Similarly, the implied share of foreign manufacturing in total manufacturing is assumed to be lower than the share observed in recent data (cf. Table 5 in the appendix). G∗ 8 N N It is assumed that ιN = G ∗ = 0 since the manufacturing output at home YN = YN and abroad is observable, while the output of the foreign nontradeable sector is not observable. For the same reason shocks to foreign services technology are excluded. 9 Note that the empirical results in Hamilton (2009) concern price elasticity of gasoline demand. Backus and Crucini (2000) study a Cobb-Douglas production function including labour and a CES bundle of oil and capital and assume a value of 0.09 for the elasticity of substition between oil and capital. They argue that this assumption is reasonable for an investigation focusing on business cycles in contrast to long-run phenonomena. Here it is abstracted from the use of capital in order to keep the analysis simple, however, the argument in favour of a rather low elasticity of substituion of oil in the production function is similar.

13

ticity between home and foreign manufactured as well as the parameters shaping the Taylor rule are identical in the two regions. The prior distributions are specified to be relatively wide so that the likelihood is allowed to have a strong weight, especially for the autocorrelation parameters of the exogenous state variables and the standard deviations of the shocks (cf. Table 1 where all the distributional assumptions are listed). The posterior mean is used for the remaining calibration of the model.10 The estimation results are in line with what is typically found in the international macroeconomics literature. The elasticity between services (nontradeables) and manufacturing (tradeables) is below one as expected (φ = 0.14), whereas the elasticity between home and foreign manufactured goods is found to be higher than one (θ = 1.36). In line with intuition, I find for the Calvo parameters a higher nominal rigidity in the services sector compared to the manufacturing sector (ϕN = 0.59, ϕT = 0.32). The obtained values imply that prices in the services sector are changed roughly between every second to third quarter, while prices in the manufacturing sector are changed roughly every first to second quarter. The estimated policy coefficients are in the range of what is usually reported in other contributions (νr = 0.60, κπ = 1.30, κy = 0.76). All estimated parameters are reported in Table 1. 10

When comparing the DSGE model of this study to the SVAR model of Kilian (2009) it can be noted that both models consists of a supply and demand block of the global oil market (with a more complex demand structure in the DSGE model). The DSGE model does, however, not incorporate what Kilian identifies as oil-market specific demand shocks in a third equation. As Ireland (2004) states, DSGE models are deeply grounded in economic theory, therefore not as flexible applicable to econometric methods such as (S)VAR times series models. He suggests the inclusion of measurement errors in order to allow for movements in the data that can not be explained by the DSGE model but would be included in VAR models. I therefore include a measurement error to the observed real price of oil. This allows a comparison with SVAR models like the one of Kilian on the one hand and improves the estimation of the deep model parameters on the other hand.

14

Table 1: Structural Estimation of Parameter Distributions, Prior and Posterior

Description

Prior

Posterior Posterior

distribution

mean

(Mean, SD) Elasticity of substitution between tradeable

φ

conf. interval

Beta (0.5,0.25)

0.14

(0.01, 0.27)

Normal (1.5,0.5)

1.36

(0.86, 1.86)

Uniform (0.5,0.29)

0.32

(0.24, 0.40)

Uniform (0.5,0.29)

0.59

(0.53, 0.65)

(manufacturing) and nontradeable (services) goods (home) Elasticity of substitution between home and

θ

foreign tradeable (manufacturing) goods (home) ϕT

Calvo parameter in the tradeable (manufacturing) sector (home)

ϕN

Calvo parameter in the non-tradeable (service) sector (home)

νr

Interest rate smoothing (home)

Uniform (0.50, 0.29)

0.60

(0.52, 0.67)

κπ

Taylor rule coefficient for inflation (home)

Gamma (1.5, 0.1)

1.30

(1.16, 1.43)

κy

Taylor rule coefficient for output growth (home)

Normal (0.8, 0.2)

0.76

(0.59, 0.92)

ρGT

Autocorrelation manufacturing demand shock

Beta (0.7,0.1)

0.87

(0.82, 0.91)

(home) ξGT

Stdv:

Beta(0.1,0.04)

0.196

(0.162, 0.231)

ρ∗GT

Autocorrelation manufacturing demand shock

Beta (0.7,0.1)

0.92

(0.89, 0.95)

(foreign) ∗ ξGT

Stdv:

Beta(0.1,0.04)

0.171

(0.151, 0.191)

ρZ T

Autocorrelation productivity shock

Beta (0.7,0.1)

0.93

(0.89, 0.96)

manufacturing (home) ξZT

Stdv:

Beta(0.008,0.005)

0.010

(0.009, 0.012)

ρZ N

Autocorrelation productivity shock services

Beta (0.7,0.1)

0.95

(0.92, 0.98)

(home) ξZN

Stdv:

Beta(0.008,0.005)

0.009

(0.008, 0.011)

ρZ ∗

Autocorrelation productivity shock

Beta (0.7,0.1)

0.72

(0.56, 0.88)

(0.001, 0.007)

T

manufacturing (foreign) ξZ ∗

Stdv:

Beta(0.008,0.005)

0.004

ρψ

Autocorrelation preference shock (home)

Beta (0.7,0.1)

0.93

(0.90, 0.96)

ξψ

Stdv:

Beta(0.01,0.005)

0.011

(0.008, 0.013)

ρ∗ψ

Autocorrelation preference shock (foreign)

Beta (0.7,0.1)

0.71

(0.55, 0.87)

∗ ξψ

Stdv:

Beta(0.01,0.005)

0.007

(0.002, 0.010)

ρos

Autocorrelation oil supply shock (global)

Beta (0.7,0.1)

0.87

(0.82, 0.91)

ξ os

Stdv:

Beta(0.01,0.005)

0.022

(0.020, 0.024)

ξr

Monetary policy shock (home), stdv:

Gamma(0.004,0.002)

0.004

(0.003, 0.005)

ξr∗

Monetary policy shock (foreign), stdv:

Gamma(0.004,0.002)

0.003

(0.001, 0.004)

T

15

3.4

Comparison of the Estimated Model with Results of Kilian (2009, AER)

The model includes 10 structural shocks in order to explain 8 observable variables. The reasonable values of the posterior means indicate that the model does very well in explaining the data and in particular US sectoral outcomes. In the following it is also examined how the imputed shock structure of the estimated DSGE model compares with the shocks of the much more parsimonious SVAR model of Kilian (2009). This allows for evaluating the external validity of the imputed shocks of the model. The basic identification approach of Kilian (2009) is a three variable, three equation model of the global oil market. The SVAR model can be interpreted as a dynamic simultaneous equation model in which the oil market is described by a supply equation, an equation describing oil factor demand and an equation that captures residual price movements respectively price movements that are caused by oil-market specific demand. The first two equations of Kilian’s model correspond to the oil endowment and the oil market equilibrium condition of the DSGE model as described in section 2, so the two models can be compared to each other. Note that Kilian (2009) uses the same oil market data as in this study (although with a different transformation) but he employs a different index for measuring real economic activity (cf. discussion above). The comparison is reported in Figure 1.11 One can observe that the imputed shocks from the estimated DSGE model do well compare to the shocks obtained with the SVAR. The annualized oil production shocks from the SVAR are positively correlated with their annualized counterparts of the DSGE model (with a correlation above 0.6). Also, the annualized global activity shocks are positively correlated with the annualized sum of foreign and domestic shocks to manufacturing demand (with a correlation 11

The shock to the supply equation in the SVAR has a natural counterpart in the DSGE model, namely the shock to the global production of oil. There is more than one shock in DSGE model that drives factor demand and the focus is restricted on the home and foreign manufacturing demand shocks. The SVAR is estimated with monthly data from 1974:1 to 2007:12.

16

of 0.38 which is high given that different data is used for identifying global industrial activity). There are still some small but notable differences in the series. The DSGE model predicts, for example, a longer uninterrupted series of positive shocks to global economic activity between 2003 and 2007 compared to the SVAR.

Figure 1: Comparison of the estimated shocks of the DSGE model with shocks obtained by the SVAR approach of Kilian (2009). The first panel compares the oil supply shock in the SVAR with the oil supply shock in the DSGE model and the second panel compares the shock to global real economic activity in the SVAR with the sum of shocks to foreign and domestic manufacturing demand in the DSGE model (expressed in terms of disturbances to the respective demand function).

4

The Trade Channel, Sectoral Dynamics and the Transmission of Oil-Related Shocks

It is of particular interest to what extent a rise in global activity that affects global oil prices is differently related to the dynamics in the trading and oil-intensive manufacturing sector compared to the non-trading and less oil-intensive services sector. The estimation of the DSGE model with several shocks allows to identify those shocks that are best able to explain the data. In the historical decomposition (cf. below) it will turn out that the shocks to domestic and foreign manufacturing demand are found to be a key demand-force for global oil prices. For the sake of readability the 17

focus is on those shocks. A comparison with US technology shocks – which are important shocks to economic activity in the US – clarifies why the latter ones are not well capturing the observed demand-type dynamics in the global oil market. The effects of a shock to the global oil endowment – that is also found to be quantitatively important for the oil price – are also discussed. 10-3

2

0.04

10-4

4

2 1

0.02 0

0

-2

0

-4 -1

-0.02 -6

-2 0

2

4

6

8

10

12

14

16

18

-0.04 20

-8 0

2

4

6

8

10

12

14

16

18

20

Figure 2: Impulse response function for a shock to the demand of foreign manufacturing (ξGT ∗ ).

Figure 2 reports a one standard deviation shock to the foreign demand for manufacturing. This foreign demand shock triggers a large increase in the price of oil and rising US net exports as well as a rising US current account. Manufacturing output in the US goes up which results in higher labour and oil demand in this sector. The output rise in manufacturing stems solely from foreign demand because aggregate consumption of the domestic agents goes down. This has contractionary effects for the services sector. Domestic wages rise on impact but then go down because of increasing labour supply. The marginal cost therefore rise in both domestic sectors on impact but stay at a high level only in the manufacturing sector which has a high oil intensity. Notably, inflation remains for many periods at a positive level in both sectors. The optimizing firms evaluate the marginal cost in terms of the price of the sector specific goods relative to the consumer price index. The relative price of services goes down which is the reason for why also firms in the services sector are adjusting prices upwards. Monetary policy is responding to rising aggregate inflation by raising interest rates consistent with the standard Taylor rule response. The shock to foreign demand for manufacturing has converse effects to 18

output of domestic manufacturing and services, namely a boom in the tradeable manufacturing sector but a contraction in the nontradeable services sector. The US trade balance goes up and labour is allocated from the services sector to the manufacturing sector. The rising nominal interest rate is amplifying the contraction in the services sector. 10-3 16

10-3

8

10-3

1

0.5

0 4

8 -0.5

-1 0

0 -1.5

-2

-4 0

2

4

6

8

10

12

14

16

18

-8 20

-2.5 0

2

4

6

8

10

12

14

16

18

20

Figure 3: Impulse response function for a shock to the demand of home manufacturing (ξGT ).

Figure 3 reports the impulse response functions for a one standard deviation shock to the domestic demand for manufacturing. It should be noted that in the case of such a demand shock infinitely lived Ricardian agents suffer a negative wealth shock. Agents reduce their consumption and supply more labour. The overall US consumption and the demand for is still positively affected by the demand shock but output in the services sector drops persistently. The demand shocks also leads to a deterioration in the current account and increasing net imports of foreign manufacturing goods. This implies a higher production of oil-intensive manufacturing goods also abroad and on aggregate the global oil price goes up. The shock leads again to a situation where domestic manufacturing output increases, services output contracts while global oil prices go up. Aggregate inflation and output growth increase and the central bank responds by raising the nominal interest rate. To summarize, shocks to home and foreign demand for manufacturing can rationalize periods of rising global oil prices and an increasing gap between output in domestic manufacturing and services, rising aggregate inflation and nominal interest rate, which is found to be an important pattern in the data (cf. below). 19

10-3 8.5

10-3

8.5

10-3

4 3.5

4.25

4.25 3

0

0

2.5 2

-4.25

-4.25 1.5

-8.5

-8.5

1 0.5

-12.75

-12.75 0

-17 0

2

4

6

8

10

12

14

16

18

-17 20

-0.5 0

2

4

6

8

10

12

14

16

18

20

Figure 4: Impulse response function for a home manufacturing technology shock (ξZT ).

It remains to discuss the role of technology shocks. The effect of a technology shock on global oil prices is a priori ambiguous. Ceteris paribus, a shock to total factor productivity in each sector lowers the demand for crude oil of this sector. At the same time, the demand in this sector goes up because of lower relative prices. This income effect increases the factor demand for crude oil. The impulse response function in Figure 4 shows that in the case of a manufacturing technology shock the first effect plays a dominant role, i.e. oil prices go down. Output goes up in both sectors while inflation goes down and the domestic central bank responds to the shock with an interest rate cut. It should be noted that the dynamics of the services technology shock are different with respect to the response of the oil price. In this case, the income effect dominates and oil prices go up modestly. This shock describes a situation in which output in both sectors rises when oil prices go up. The results of the historical decomposition indicate that the technology shocks play an important role in explaining the dynamics of the US economy, however, they affect the global oil price only to a relatively small extent. (cf. below). Figure 5 reports the effects of a one-standard deviation shock to the global oil endowment. This shock plays a very prominent role in most of the literature on the relationship between real oil prices and the macroeconomy. As expected, an increase in the global production of oil leads to a – indeed sizeable – decrease in the price of oil. Importantly, that change is affecting both sectors with the same sign and the oil-intensive manufacturing sector is not very differently affected compared to the services sector. The manu-

20

10-3

7

0.14

3.5

0.07

10-3

3

2.5

2 0

0 1.5

-3.5

-0.07

-7

-0.14

-10.5

-0.21

1

0.5

-14 0

2

4

6

8

10

12

14

16

18

-0.28 20

0

-0.5 0

2

4

6

8

10

12

14

16

18

20

Figure 5: Impulse response function for a shock ξos to the global production of crude oil.

facturing sector has a higher oil-intensity but is also facing lower nominal rigidity. It is interesting to note that the analysis below reveals that the oil-production shock can not explain important changes in the price of oil, like for example the 1979 surge or the dynamics in 2003-2008 but still it plays an important role in other interesting episodes like for example for persistent high oil prices in the mid-1980s.

5

Historical Dynamics of the Oil Price, Sectoral Patterns and the Aggregate Economy

In this section I conduct variance decomposition of the shocks of the model in order to discuss the transmission mechanisms during interesting historical episodes in the global oil market. The graphs describe how in every point in time real prices and quantities are explained by the various exogenous disturbances. The historical decomposition identifies (US and foreign) manufacturing demand shocks as a key driver of global real oil prices. These shocks rationalize changing oil prices and diverging output in the tradeable manufacturing and the nontradeable services sector in important episodes like the oil price rise in 2003-2007, the pattern around the 2001 recession or the fall in oil prices in 1986. I first want to focus on the prominently discussed episodes of 2003-2007 and the time around the double dip recessions in the early 80’s. 21

The 2003-2007 Surge in Oil Prices The episode from 2003 until 2007 had been characterized by surging global real oil prices and global activity. In the US, sectoral outputs were at or below their trend levels but the manufacturing sector was still doing better than the services sector. Aggregate inflation and the US nominal interest rates increased steadily (cf. Figure 6).

Figure 6: Dynamics in 2003-2008; for variables with a trend the values can be interpreted as percentage deviations from the trend.

The estimation procedure identifies manufacturing demand shocks (in this periods especially originating in foreign) as the main explanation for this pattern. This is in line with anecdotal evidence for this episode which is characterized by industrialization in Asia. The interesting finding is that the demand shocks almost perfectly explain the gap between manufacturing and services in the US as well as the rise in global oil prices. The intuition of this result is that the US manufacturing sector benefits from additional global demand because of higher exports to the rest of the world. As described above, aggregate inflation goes up and labour is allocated from the services sector to the manufacturing sector. These results oppose the common view that oil-intensive industries are hit first by higher oil prices because of their cost structure. The trade channel as operative here has 22

the interesting implication of a booming oil-intensive manufacturing sector in times of high oil prices. It should be noted that the manufacturing sector is largely positively contributing to aggregate US output during that time but the contraction in the services sector is counteracting this effect. In sum, the net effect of the manufacturing demand shock on US aggregate output is positive but small. There is, nevertheless, a large effect of the shock on aggregate US inflation. This implies the interesting situation of high aggregate inflationary pressure without much changes in aggregate output. The manufacturing demand shocks are not the sole explanation of the observed contraction in services during this episode. US specific technology shocks do also to contribute to the slump. These shocks do not affect global oil prices importantly but lead to a sizeable contraction in both sectors. This is in line with results in the literature finding output growth in the US to have slowed down already prior to the Great Recession. Fernald (2014) argues that total factor productivity already decreased in the mid 2000s because of factors originating in the IT-intensive industries of the economy.

The Volcker Recessions The historical decomposition in Figure 7 reveals that the episode before and after the double dip recessions in the early 80’s has interesting similarities with 2003-2007. Output in the US manufacturing sector as well as global oil prices are positively affected by shocks to (US and foreign) demand for manufacturing but the US services sector contracts. Aggregate inflation in the US economy increases and the central bank is responding by raising the nominal interest rate. The last result sheds light on the issue of the role of the central bank for the oil-price / macroeconomy relationship. Bernanke et al. (1997) and Bernanke et al. (2004) employ VAR models and find for this episode – in which Paul Volcker was chairman of the Fed – that the central bank tightened interest rates as a systematic response to rising oil price which as a consequence amplified potential recessionary effects of oil prices. The important underlying assumption in these studies is that oil prices are driven 23

Figure 7: Dynamics around the Volcker Recessions; for variables with a trend the values can be interpreted as percentage deviations from the trend.

by oil supply shocks and not by shocks that are related to the US economy. I find as a result that this assumption should be challenged because oil production was high during this episode before 1980 and did not contribute to domestic inflation, so the observed monetary policy response tightening is not related to negative oil supply shocks.12 The historical decomposition shows that a large fraction (5-8 basis points) of the high central bank rate at that time can be rationalized by a systematic response to aggregate inflation and output growth caused by manufacturing demand shocks. High oil prices were a symptom of high global manufacturing production but did not directly attract the attention of the central bank. One should note that the central bank response to aggregate inflation and output fits the manufacturing sector but it does not fit the contracting services sector. In that sense a systematic response to aggregate inflation and output has converse effects because it can not counteract the sectoral 12

The literature often discusses supply disruption caused by the Iranian Revolution in 1978/79 as an important event in global oil markets. I find for these years that global production was still high and that only after oil prices started rising in 1979 negative oil supply shocks contributed to high oil prices from 1981 on (a time were the Iran-Iraq war caused oil-disruption), cf. Figure 7.

24

output divergence in such a two-speed economy. It remains to discuss the effect of oil-related shocks for the recessions themselves. I find that in the second and larger recession, oil-related shocks contributed to roughly a 1 % contraction in real GDP. The contraction partly has its source in demand-type oil related shocks (a bust in manufacturing demand that followed the described boom) but in this particular period after 1980 mainly in oil supply shocks since lowering oil endowment started to pressure oil prices from the first period of 1980 on. The effect of all (demand and supply-type) oil-related contractionary sources is large but nevertheless modest compared to other contractionary effects of this time (which are not related to oil prices in a quantitative important way). Kilian and Vigfusson (2014) employ a structural VAR and find for the time of the Volcker recessions that a positive oil price shock has within two years a cumulative contractionary effect on output of around 1 %, which is in line with my findings.

Other Interesting Episodes There were many other interesting episodes of mutually interacting oil prices and macroeconomic dynamics. Focusing on the distinction between manufacturing and services can, for example, shed light on the events before and after the 2001 recession. The historical decomposition reveals that again the up and downswings in oil prices correlate to the widening and closing of the gap in the output cycle between manufacturing and services.13 The historical decomposition indicates that, prior to the recession, shocks to (especially) home and foreign manufacturing demand contributed to the large positive cyclical gap between manufacturing and services in the US and to rising oil prices. The period is still different to the two pre-recession episodes in the 70’s and in 2003-2007 since the large cyclical gap before 2001 is not solely explained by manufacturing demand shocks but to a larger extent by shocks to US manufacturing technology (which do not much affect oil prices). This finding is consistent with the evidence reported in Basu et al. (2001) who find that in that in the 1990s positive technology shocks affected especially the durable manufacturing sector of the US – much more 13

The detailed historical decomposition can be found in the appendix.

25

than other sectors of the economy. Interestingly, I find that manufacturing technology was not the reason for the recession in 2001, instead it was a combination of decreasing demand and negative technology shocks in the services sector. In the period of the 2001 recession I find that oil production was actually increasing and these oil supply shocks affected oil prices negatively and US aggregate output positively. Between 2001 and 2003, especially negative demand shocks explain oil prices with the exception of 2002 when negative oil supply shocks contribute positively. A possible explanation for this finding could be the Venezuelan crisis of 2002. This shock does, however, not affect sectoral or aggregate macroeconomic dynamics in the US. I find no effect of the 2003 Iraq war on oil prices. Like the other recessionary episodes, also the 1990 recession was preceded by rising oil prices, in this case by a short-lasting hike. It is important to note that in the time before and after the recession the US manufacturing output cycle was below the US services cycle, which is different compared to all other recessions in the sample. Important for the focus of this study is that the negative gap between the cycle in manufacturing and services at that time does still relate to the dynamics in the global real oil price. Despite the short-lasting hike that preceded the recession, the real oil price was at a relatively low level. The historical decomposition indicates that especially weak US manufacturing demand contributed to the low level that lasted long after the recession. Negative oil supply shocks were in some periods – that coincide with the war in the Persian gulf – contributing to the hike in oil prices, however, this hike did not much affect sectoral patterns or the aggregate US economy. The large fall in oil prices in 1986 is also worth to discuss. The episode prior to 1986 had been characterized by a persistent fall in manufacturing output, at the same time the services sector did relatively well. The central bank lowered interest rates in this time of falling (sectoral and aggregate) inflation. The sectoral gap narrowed in the first quarter of 1987 at the same time when oil prices started to rise again The historical decomposition indicates that negative home and foreign manufacturing demand shocks can rationalize this pattern (cf. Figure 8). This finding has the important

26

Figure 8: Falling oil prices in 1986; for variables with a trend the values can be interpreted as percentage deviations from the trend.

implication that shocks to manufacturing demand can explain episodes of rising as well as falling real oil prices. In the period of falling oil prices in 1986 as well as in other episodes, for example in 1979 or between 2003-2008 some fraction of the variation in oil prices remains unexplained, which is captured by the measurement error. The empirical literature gives some hints for this result. Kilian and Murphy (2014) study the presence of speculative pressures in the market for oil to identify unobserved shifts in expectations about oil demand. They find that such shifts played a role in the surge in prices in 1979 or the fall in 1986. In these episodes the assumed measurement error in the oil price equation of the DSGE model coincides with their evidence for shifts in expectations. There are other episodes where the DSGE model predicts uncaptured variation, most prominently in 2003-2008. Kilian and Murphy (2014) do not find speculative pressure in this episode. On the contrary, Juvenal and Petrella (2015) argue that financialization in commodity markets contributed importantly to the rise in oil prices at this time. While this debate is outside the focus of this study, it should be noted that there is no evidence for an important relation between the unidentified variation

27

in oil prices and the here discussed trade channel.

6

What Happened During 2008-2016?

In the previous section it was emphasized that the observed dynamics in global oil markets are related to the divergence between output of the manufacturing and the services sector in the US. The model was estimated up until the last quarter of 2007 and following periods are not included in the analysis above. In this section, I want to focus on the years 2008-2016 by conducting forecasting and in-sample prediction exercises. The main aim of this exercises is to study the dynamics of this more recent episode and also to check the explanatory power of the model. As a first exercise, I employ the estimated model from the previous section and conduct an unconditional forecasting exercise (cf. Figure 9). In particular I study how the system studied above evolves from the last observed point (2007:Q4) onward without using any data observed after this point (i.e. the projection does not incorporate effects of the 2008 financial crisis). 1

0.02 0

0.8

-0.02 0.6 -0.04 0.4

-0.06 -0.08

0.2

-0.1

0

-0.12 -0.2 -0.14 -0.4

-0.6 2007Q4

Real Oil Price (observed) Forecasted Path

2009Q2

2010Q4

2012Q2

2013Q4

Wedge btw. Manufacturing and Services Forecasted Path

-0.16

2015Q2

-0.18 2007Q4

2016Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

2016Q4

0.01 0.005

-0.02

0 -0.04 -0.005 -0.06

-0.01

US GDP (observed) Forecasted Path

-0.015 -0.08 -0.02 -0.1

-0.025 US CPI Inflation (observed) Forecasted Path

-0.03 -0.035 2007Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

-0.12

2016Q4

2007Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

2016Q4

Figure 9: Unconditional Forecasts given the estimated model for data until 2007:Q4. The predicted series are expressed in percentage deviation from the trend or the sample mean in case of stationary variables.

28

The most notable finding is that the forecasted medium term evolution of the global real oil price is similar to the actually observed path. The model predicts a persistent fall in oil prices from around 100 US$ in 2007 to around half the price in 2015. The reason for this prediction is that high global manufacturing demand was explaining the rise in oil prices before 2007. The forecasted path explains a situation in which shocks fade out and the system is going back to its steady state. This is consistent with the narrative that industrialization in Asia is cooling down at some point. The fading out of global manufacturing demand has negative effects on US manufacturing output which is forecasted to persistently decrease after 2007. Negative technology shocks – that are not related to dynamics in global oil markets – exaggerate this pattern. The US services output is, nevertheless, forecasted to slowly increase after 2007 and the gap between manufacturing and services turns negative. The contraction in manufacturing and the slow reversal in services do imply very small growth rates for aggregate output. Aggregate output is around three percentage points below steady state in the beginning of 2008 and improves only by two percentage points until 2015. This trajectory is unconditional on the crisis occurring in 2008, therefore it is not surprising to find for 2015 an around 10 percentage point deviation from steady state in the actual data compared to a one percentage points deviation in the unconditional forecast path. Interestingly, the forecasting exercise predicts aggregate inflation rates that are close to the observed ones. The results indicate that a low-growth, low-inflation period with low interest rates would have characterized the years until 2015 even without a financial crisis happening in 2008. The next exercise goes one step further and gives a forecast for 2008:Q12016:Q4 conditional on having observed the large output contraction happening in 2008-2009. For doing this, I use observed values for global industrial activity and US services output for 2008:Q1 until 2010:Q2 and compute forecasts for all variables from 2008:Q1 until 2016:Q4.14 I allow shocks to foreign demand for manufacturing and US manufacturing technology shocks to explain the conditional path during 2008:Q1-2010:Q2. The reason for this choice is that these shocks are found to be quantitatively 14

The data source is the same as above. I use and report recursively linearly detrended data until 2016:Q4.

29

important before 2008. In this benchmark exercise global oil production data is not used. 1.2

0.05 Wedge btw. Manufacturing and Services Forecasted Path Forecasted Path (including oil supply data)

1 0.8

0

0.6 0.4

-0.05

0.2 0

-0.1

-0.2 Real Oil Price Forecasted Path Forecasted Path (including oil supply data) Forecasted path (no nontradeable sector)

-0.4 -0.6 -0.8 2007Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

-0.15

-0.2 2007Q4

2016Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

2016Q4

0

0.08 CPI Inflation Forecasted Path Forecasted Path (including oil supply data) Forecasted Path (no nontradeable sector)

0.06

-0.02 -0.04 -0.06

0.04

-0.08 0.02 -0.1 -0.12

0

-0.14

US GDP Forecasted Path Forecasted Path (including oil supply data) Forecasted Path (no nontradeable sector)

-0.02 -0.16 -0.04 2007Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

-0.18 2007Q4

2016Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

2016Q4

Figure 10: Forecast conditional on observations for 2008:Q1 to 2010:Q2 (vertical dotted line). Observations include in the benchmark exercise the global industrial activity index and US services output, in the first extension additionally oil supply data, and for the second extension (model without nontradeable sector) global industrial activity and US GDP. The predicted series are expressed in percentage deviation from the trend or the sample mean in case of stationary variables

I report two extension of the benchmark exercise. In a first extension I additionally use observed oil supply data and additionally allow for oil supply shocks. By doing so it can be studied if oil supply data can improve the forecast compared to the case where only data on services output and global industrial activity is used. In a second extension to the benchmark exercise I employ a similar model that does not include a nontradeable sector and is estimated with aggregate US data.15 This allows to check if incorporating a nontradeable sector has advantages with respect to explanatory power. In this extension I allow for a shock to domestic technology and to foreign demand (for the single tradeable good). Figure 10 reports all the results. The benchmark forecasting exercise does 15

In this model the US home bias in consumption is set to δ = 0.15 such that the calibrated model is comparable with the benchmark model. All other parameters – if they exist and are not estimated – remain with the same calibration. Additional to the mentioned data sources, I use data for US GDP and CPI inflation to estimate the model.

30

very well predict the observed real oil price in 2008/2009. This implies that with the model studied in this paper, the information about global activity and US services output suffices to capture the – demand driven – dynamics in real oil price. The manufacturing demand shock identifies the importance of the trade channel for the gap between US manufacturing and services in 2008/09. This is consistent with anecdotal evidence for these two years when a large fall in global trade heavily affected US manufacturing (for example the US automobile industry). The services sector did relatively better because it was not much affected by the fall in global trade, at the same time oil prices fell. This pattern is similar to other episodes discussed above. The technology shock still played a role in driving both sectors down and also this shock affects the manufacturing sector more than the services sector, however, it is not related to falling oil prices. The forecast after 2010 compares reasonably well to the observed data. Real oil prices are forecasted to go down in the medium run and output growth and inflation is predicted to be low as it happened in reality. The observed gap between manufacturing and services is almost perfectly matched between 2010 and 2013. The benchmark model with a nontradeable sector is producing a better forecast compared to the the extension including oil supply shocks. It is also outperforming the model with only one aggregate (tradeable) sector. In the model with only one sector, the global oil prices is predicted to fluctuate much more than was actually observed before 2010 and afterwards remains almost constant at a very low level. This indicates that the model with only one sector can not well explain oil demand dynamics. In order to further examine the explanatory power of the benchmark model and shed some light on the last observations in the sample I conduct the same in-sample prediction exercise as above conditional on observed global industrial activity and US services for the whole time span from 2008:Q1 until 2016:Q4.16 The benchmark exercise can fit real oil prices reasonably well (cf. Figure 11) and does also in this case outperform the prediction which is based on 16

In this exercise the predicted trajectories of US GDP and the sectoral wedge between US manufacturing and services are partly determined by observed variables directly and therefore not of particular interest.

31

1.2

0.08 CPI Inflation Predicted Path Predicted Path (including oil supply data) Predicted Path (no nontradeable sector)

1 0.06

0.8 0.6

0.04

0.4 0.2

0.02

0 0

-0.2 Real Oil Price Predicted Path Predicted Path (including oil supply data) Predicted Path (no nontradeable sector)

-0.4 -0.6 -0.8 2007Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

-0.02

-0.04 2007Q4

2016Q4

2009Q2

2010Q4

2012Q2

2013Q4

2015Q2

2016Q4

Figure 11: Prediction conditional on observations for 2008:Q1 to 2016:Q4. Observations include in the benchmark exercise the global industrial activity index and US services output, in the first extension (including oil supply shocks) additionally oil supply data, and for the second extension (model without nontradeable sector) global industrial activity and US GDP. The predicted series are expressed in percentage deviation from the trend or the sample mean in case of stationary variables

the model without a nontradeable sector. The results indicate that after 2013 demand factors contributed to decreasing oil prices. It is notable that the observed wedge between US manufacturing and services is decreasing during this episode (cf. Figure 10) and again comoving with the path of oil prices. After 2013, the oil price is, however, better predicted in-sample when oil production data is used in the benchmark model with a nontradeable sector. In particular the oil price drop in 2014 is shaped by increasing oil production. These findings are in line with the results of Baumeister and Kilian (2016) who find that slowing demand factors and high oil production contributed to the slump in oil prices in 2014/15. It should be noted that the oil supply shock does not improve the predicted path of US CPI inflation which is reasonably well matched by the benchmark model. In particular, during 2014 observed US CPI inflation is decreasing. The shocks to manufacturing activity are able to capture these dynamics in the predicted trajectory.

7

Conclusion

Understanding the causes and consequences of up and downswings in global oil prices is important for the conduct of monetary and fiscal policy. While the classical explanation of oil price movements are disruptions in global oil production, the recent literature stresses the role of demand-driven changes 32

in global oil prices and their consequences for macroeconomic outcomes in a large open economy. The trade channel plays thereby an important role. When, for example, Asian industrialization is contributing to a boom in the global business cycle, then this fuels the global oil price because of increasing factor demand but also leads to rising non-oil exports of large open economies like the US which has positive effects on their business cycle. In this paper, I argue that for studying the oil / macroeconomy relationship, one should carefully distinguish between the manufacturing sector that is exporting a lot and uses oil heavily versus the typically large services sector that is exporting very little and does not use oil intensively. I study a model of two large open and oil-dependent economies (US versus rest of the world) in which each country has a trading and oil-intensive manufacturing sector that is differently related to underlying sources of global oil price changes (such as growing Asian activity) than a relatively large but non-trading services sector. The model is estimated with a Bayesian approach and with data on global oil production and prices, output and inflation in the US manufacturing and the US services sector as well as a proxy for global industrial production for the time span 1974:Q1 until 2007:Q4. I find as a main result that the global oil price is driven by expanding manufacturing (at home or abroad) and is related to the gap between manufacturing and services in the US. Manufacturing demand shocks are identified to account for quantitative sizable changes in oil prices. Those shocks fuel oil prices and cause an increase in manufacturing as well as in domestic inflation and nominal interest rates but a contraction in services. Such a pattern is observed during important episodes, including the 20032008 surge in oil prices, the episode before the double dip recessions in the early 80’s, the episode around the 2001 recession or (with different signs) the fall of the oil price in 1986. The results indicate that the trade channel plays a crucial role in the transmission of oil-related shocks to the economy. Focusing on the role of this channel and distinguishing between manufacturing and services allows a novel view on historical events. I find that the 70’s are not so much different from the 2000’s in a sense the US economy is affected similarly by demand-driven oil price changes. While a common view is that large changes in the oil prices are bad for an economy because they affect pro33

duction in oil-intensive sectors like manufacturing, I find on the contrary that in times of rising oil prices the manufacturing sector actually benefits from global activity, whereas this is not the case for the less oil intensive services sector. I find that the US central bank raises the interest rate in times of booming global manufacturing and a rising oil price. Monetary policy does thereby not directly respond to rising oil prices but raises the interest rate because of higher aggregate inflation and output growth. This policy fits the manufacturing sector but amplifies contractionary sources for the large services sector. The central bank can not counteract the diverging pattern in the two sectors of the economy. In such a two-speed economy, there is no natural rate that fits each sector. If the policy goal is to stabilize output then one should consider a policy mix and the use of fiscal policy in order to counteract the reallocation effects between the tradeable and the nontradeable sector in times of booming global trade and oil prices.

34

References An, S. and Schorfheide, F. (2007). Bayesian analysis of dsge models. Econometric Reviews, (Volume 26, Issue 2-4). Backus, D. K. and Crucini, M. J. (2000). Oil prices and the terms of trade. Journal of International Economics, 50(1):185–213. Balke, N. S., Brown, S. P. A., and Yucel, M. K. (2010). Oil price shocks and U.S. economic activity: An international perspective. Discussion Paper dp-10-37, Resources For the Future. Basu, S., Fernald, J. G., and Shapiro, M. D. (2001). Productivity growth in the 1990s: technology, utilization, or adjustment? Carnegie-Rochester Conference Series on Public Policy, 55(1):117–165. Baumeister, C. and Kilian, L. (2016). Understanding the decline in the price of oil since june 2014. Journal of the Association of Environmental and Resource Economists, 3(1):131–158. Beidas-Strom, S. and Pescatori, A. (2014). Oil price volatility and the role of speculation. IMF Working Paper, WP/14/218. Benigno, G. and Thoenissen, C. (2003). Equilibrium exchange rates and supply-side performance*. The Economic Journal, 113(486). Benigno, P. (2009). Price stability with imperfect financial integration. Journal of Money, Credit and Banking, pages pp 121–149. Bernanke, B., Gertler, M., and Watson, M. W. (2004). Oil shocks and aggregate macroeconomic behavior: The role of monetary policy: A reply. Journal of Money, Credit, and Banking, 36(2):287–291. Bernanke, B. S., Gertler, M., and Watson, M. (1997). Systematic monetary policy and the effects of oil price shocks. Brookings Papers on Economic Activity, 28(1):91–157. Blanchard, O. J. and Gali, J. (2010). The Macroeconomic Effects of Oil Shocks: Why are the 2000s So Different from the 1970s? In International

35

Dimensions of Monteray Policy, ed. by J. Gali and M. Gertler (Chicago: University of Chicago Press), pp. 373-421. Blanchard, O. J. and Riggi, M. (2013). Why are the 2000s So Different from the 1970s? A Structural Interpretation of Changes in the Macroeconomic Effects of Oil Prices. Journal of the European Economic Association, 11(5):1032–1052. Bodenstein, M., Erceg, C. J., and Guerrieri, L. (2011). Oil shocks and external adjustment. Journal of International Economics, 83(2):168– 184. Bodenstein, M., Guerrieri, L., and Kilian, L. (2012). Monetary policy responses to oil price fluctuations. IMF Economic Review, 60(4):470– 504. Campolmi, A. (2008). Oil price shocks: Demand vs supply in a two-country model. Magyar Nemzeti Bank (the central bank of Hungary), (No. 20085). Carlstrom, C. T. and Fuerst, T. S. (2006). Oil prices, monetary policy, and counterfactual experiments. Journal of Money, Credit, and Banking, 38(7):1945–1958. Dotsey, M. and Duarte, M. (2008). Nontraded goods, market segmentation, and exchange rates. Journal of Monetary Economics, Volume 55, Issue 6:Pages 1129–1142. Fernald, J. G. (2014). Productivity and potential output before, during, and after the great recession. Federal Reserve Bank of San Francisco Working Paper Series. Ferrero, A., Gertler, M. J., and Svensson, L. E. (2010). Current Account Dynamics and Monetary Policy. In International Dimensions of Monetary Policy, ed. by J. Gali and M. Gertler (Chicago: University of Chicago Press), pp. 373-421. Hamilton, J. (2009). Understanding Crude Oil Prices. The Energy Journal, Volume 30(Number 2):179–206. 36

Hamilton, J. D. (1983). Oil and the macroeconomy since world war II. Journal of Political Economy, 91(2):228–248. Hamilton, J. D. (2011). Nonlinearities and the macroeconomic effects of oil prices. Macroeconomic Dynamics, 15(S3):364–378. Ireland, P. N. (2004). A method for taking models to the data. Journal of Economic Dynamics & Control, 28:1205–1226. Juvenal, L. and Petrella, I. (2015). Speculation in the Oil Market. Journal of Applied Econometrics, 30(4):621–649. Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply shocks in the crude oil market. The American Economic Review, 99(3):1053–1069. Kilian, L. and Hicks, B. (2013). Did Unexpectedly Strong Economic Growth Cause the Oil Price Shock of 2003-2008? Journal of Forecasting, 32(5):385–394. Kilian, L. and Murphy, D. P. (2014). The role of inventories and speculative trading in the global market for crude oil. Journal of Applied Econometrics, 29(3):454–478. Kilian, L. and Vigfusson, R. (2014). The Role of Oil Price Shocks in Causing U.S. Recessions. SSRN Scholarly Paper ID 2501512, Social Science Research Network, Rochester, NY. Leduc, S. and Sill, K. (2004). A quantitative analysis of oil-price shocks, systematic monetary policy, and economic downturns. Journal of Monetary Economics, 51(4):781–808. Lubik, T. and Schorfheide, F. (2005). A bayesian look at the new open economy macroeconomic. NBER Macroeconomics Annual, Volume 2. Natal, J.-M. (2012). Monetary policy response to oil price shocks. Journal of Money, Credit and Banking, 44(1):53–101. Rabanal, P. and Tuesta, V. (2010). Euro-dollar real exchange rate dynamics in an estimated two-country model: An assessment. Journal of Economic Dynamics and Control, 34(4):780–797. 37

Rabanal, P. and Tuesta, V. (2013). Nontradable goods and the real exchange rate. Open Economies Review, 24(3):495–535. Schmitt-Grohe, S. and Uribe, M. (2003). Closing small open economy models. Journal of International Economics, 61(1):163–185. Smets, F. and Wouters, R. (2003). An Estimated Dynamic Stochastic General Equilibrium Model of the Euro Area. Journal of the European Economic Association, 1(5):1123–1175. Stockman, A. C. and Tesar (1995). Tastes and technology in a two-country model of the business cycle: Explaining international comovements. The American Economic Review, Vol. 85, No. 1 (Mar., 1995), pp. 168-185.

38

Appendix A: Data and Graphs Data Source

Observable variable Real manufacturing output (US) Real services output (US) Quarterly inflation in manufacturing (US) Quarterly inflation in services (US) Global oil supply

Real price of oil

Nominal interest rate (US) Global level of industrial activity (level of global trade)

Data Source

Observational variable

Index of manufacturing output (US); (OECD Main Economic Indicators)

obs YT,t

Index of services output (US); (US Bureau of Economic Analysis, BEA)

obs YN,t

Quarterly growth rate of US Producer Price Index for Manufacturing (OECD Main Economic Indicators)

πTobs H,t

Quarterly growth rate of US Consumer Price Index for Services (US Bureau of Labour Statistics, BLS) World oil production (Source: US Energy Department) Refiner import price of crude oil (Source: US Energy Department) divided by US Consumer Price Index (Source: CPI US All Urban Consumers, BLS) US Treasury Bill Rate (transformed to quarterly frequency); Source: IMFInternational Financial Statistics Index of global industrial production (cf. Beidas-Strom and Pescatori (2014) and footnote 5)

Table 4: Observed variables

39

obs πN,t

obs Os,t

obs Poil,t

rtobs

REAobs t

Sectoral Oil Use and Tradeability from US Input-Output Data Input shares (USD per USD value added) Oil TransUtilities port

Sector

% GDP

Exports (% of VA)

Services Manufacturing Government Construction Agriculture Mining (w/o oil)

64.71 % 12.78 % 11.63 % 4.96 % 1.03 % 0.52 %

3.83 % 38.26 % 0.00 % 0.01 % 19.53 % 8.70 %

0.008 0.040 0.033 0.070 0.113 0.055

Transport Utilities

2.63 % 1.67 %

23.38 % 0.62 %

0.197 0.360

0.015 0.058 0.022 0.035 0.050 0.094

0.010 0.021 0.014 0.017 0.022 0.044

Table 5: The role of oil and openness for the sectors of the US economy (BEA, Input Output Tables 2005)

Detailed Shock Decomposition (ad section 4)

Real Oil Price

40

Wedge between output in US manufacturing and US services

US Nominal Interest Rate

US Inflation Rate

41

US manufacturing output

US services output

US GDP (weighted sum of sectoral output)

42