Has inflation targeting become less credible?

BIS Working Papers

No 729

 

Has inflation targeting become less credible?

by Nathan Sussman and Osnat Zohar

 

Monetary and Economic Department

June 2018

 

JEL classification: E52, E58, E31, E32

Keywords: Inflation targeting, inflation expectations, monetary policy, oil prices, anchoring, credibility

 

This publication is available on the BIS website (www.bis.org).

 

© Bank for International Settlements 2017. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISSN 1682-7678 (online)

 

Has inflation targeting become less credible?

Nathan Sussman and Osnat Zohar

 

Abstract

Beginning with the global financial crisis (2008) the correlation between crude oil prices and medium-term and forward inflation expectations increased leading to fears of their un-anchoring. Using the first principal component of commodity prices as a measure for global aggregate demand, we decompose nominal oil prices to a global demand factor and remaining factors. Using a Phillips Curve framework we find a structural change after the collapse of Lehman Brothers when inflation expectations reacted more strongly to global aggregate demand conditions embedded in oil prices. Within this framework we cannot reject the hypothesis that expectations remained anchored.

Keywords: Inflation targeting, inflation expectations, monetary policy, oil prices, anchoring, credibility

JEL Classification: E52, E58, E31, E32


1. Introduction

The sharp decline in oil prices beginning in late 2014 sparked a debate about their effect on inflation and the world economy [e.g., Chen et al. (2015); Baumeister and Kilian (2016a,b)]. This decline lowered inflation in the short run and in some cases resulted in negative inflation [IMF (2016)]. More surprisingly, data from the USA, France, UK and Israel show that oil prices have a strong correlation with inflation expectations for the medium-term, as mea­sured by five-year breakeven inflation rates, and more recently with five-years to five-years forward inflation expectations (Figure 1). Before the global financial crisis (GFC), with the exception of the UK, this correlation was weaker and expectations were firmly anchored at two percent. However, from the onset of the GFC the correlation is quite high, suggesting that expectations for the five-year horizon became less anchored with respect to the inflation target. While this phenomenon is more visible in medium-term inflation expectations, since 2014 we can observe a similar pattern with respect to longer term inflation expectations, namely the five-year five-year forward breakeven rates. A concern raised by central bankers was that the increased correlation of breakeven inflation rates with oil prices may indicate an erosion of the anchoring of expectations [e.g., IMF (2016)]. These developments may signal a decline in either the effectiveness, appropriateness or credibility of the inflation targeting monetary regime and questions conclusions reached about the credibility of this regime and its effect on the anchoring of inflation expectations [Gurkaynak et al. (2010); Beechey et al. (2011)].

Our main contribution is to test for un-anchoring of inflation expectations using a struc­tural framework based on a novel global Phillips curve. We construct measures of global demand shocks - that we extract from commodity prices, global inflation and monetary policy. Using this approach we find a structural change in the effect of global demand on inflation expectations following the onset of the GFC. However, this does not necessarily imply that expectations became unanchored. In fact, we cannot reject the hypothesis that they remained anchored.

The observed correlation between oil prices and medium-term inflation expectations that alarmed policy makers and motivated this study is puzzling since we do not expect a cor­relation between (expected) rates of change in the CPI in the medium term and levels of oil prices. However, Killian, in numerous studies [Barsky and Kilian (2001, 2004); Kilian (2008a, 2009); Kilian and Hicks (2013)]), already showed that real oil prices convey infor­mation about global economic activity and therefore could be related to inflation.

We extend Killian’s approach by focusing on the increase (decrease) in the aggregate price level, and hence inflation expectations, caused by changes to aggregate demand or supply. We begin by identifying the component of oil price changes that is affected by global aggregate demand. We exploit the fact that a large number of commodity contracts are traded in financial markets. While each commodity is affected by idiosyncratic supply and demand shocks, they are also affected by common “global aggregate demand” shocks. Since idiosyncratic changes in the price of one commodity may affect other prices in differ­ent directions (depending on substitution and income effects), a factor that can move the prices of all commodities in the same direction is global aggregate demand. We exploit

Image bwf

the strong co movement of commodity prices to identify global aggregate demand as their first principal component. The advantage of using a common factor of commodity prices to extract information about global aggregate demand is that these prices are derived from almost perfect markets: they are standardized goods, traded in thick markets, and there is global full-information of their prices. Figure 2 shows that the common factor we extract

Image sxx

from commodity prices is strongly correlated (correlation coefficients higher than 0.8, see section 2.1.2) with measures of the global output gap.

Our measure of global aggregate demand, embedded in oil prices, is largely responsible for the increase in the correlation between them and breakeven inflation expectations. To rule out that the correlation did not increase due to oil specific effects, we also include in our estimation the residual from a regression of the change in oil prices on our measure of global demand. We further instrument the residual with specific variables affecting oil and energy prices idiosyncratically, namely OPEC's strategic behavior and shocks to oil demand caused by the weather. We construct a novel proxy for OPEC's behavior by using a tally of articles from the London Times. We examine articles that mention OPEC and classify them by the sentiment arising from the text. Our proxy is constructed as the net number of articles suggesting OPEC is expanding supply, minus the number of articles indicating supply reduction. We also use temperature variables from five continents to capture shocks to the demand for oil arising from anomalous weather conditions. Using these factors as instruments supports our hypothesis that the increase in the correlation between oil prices and inflation expectations was largely due to changes in the correlation between global aggregate demand and expected inflation.

We find an increased volatility of breakeven expectations due to a structural change in the correlation between them and global demand. To address the question whether our results imply an un-anchoring of inflation expectations, we test for un-anchoring using a structural model with rational expectations. The structural framework, together with our decomposi­tion of oil prices, allows us to identify the channel by which inflation expectations react more strongly to the global demand factor embedded in oil prices. We estimate a reduced form ’global’ Phillips curve, taking a similar approach to Ciccareli and Mojon (2010) and Diebold et al. (2008) who used global principal component analysis. We exploit the fact that all advanced economies are part of the monetary regime referred to as ’inflation targeting’. The USA, the ECB and, to some extent, the UK are perceived as the global anchors of this regime and therefore, there is a common factor in medium-long term inflation expectations for these economies. This allows us to compute a global measure of expected inflation using the first principle component of inflation expectations in these economies. Similarly we construct measures of global inflation and global monetary policy. This method removes idiosyncratic shocks to expected inflation and bond markets from which these expectations are extracted (for example the shock of Brexit on the London capital market).

We estimate a reduced form global Phillips curves (with and without assuming a mon­etary policy rule) using mainly market based five year breakeven expected inflation rate, but also one year ahead expectation based on consumer surveys. We find, again, that the increase in the correlation between oil prices and expected inflation, following the onset of the crisis, is mainly due to the increased effect of global aggregate demand on global inflation expectations. Within the assumptions of the model we used, we can reject the hypothesis of un-anchoring and attribute our findings to an increase in the slope of the global Phillips curve during the period following the onset of the global financial crisis.

The implication of our findings for policy makers is that the increased volatility of medium term inflation expectations does not necessarily imply un-anchoring. Nevertheless, central bankers are concerned that the increased volatility of medium term inflation expectations may cause, over time, a deterioration of the credibility of inflation targeting. In fact, since 2014 we observe a decline in the long term inflation expectations - the five-year to five- year breakeven rates and inflation swaps. This development may indicate en erosion in the public belief in the ability of central banks to bring inflation back to target in the longer run, either because monetary may be less affective around the effective lower bound, or that central bankers may be less willing to use monetary policy tools because of potential macro-prudential concerns. Another possibility, recently raised by Morris and Shin (2018), is that central bankers’ communication of perhaps, according to our findings, unjust fears of un-anchoring have led to the decline in long term inflation expectations.

While outside the scope of this paper, we found that there is a possibility that the slope of the global Phillips curve increased after the global financial crisis. Our findings suggest that the volatility of inflation and inflation expectations may increase and that there are fewer frictions in the global economy and particularly in prices. This implies that economists and policy makers should consider the possibility of this scenario and its implications for the monetary rule and the models used to support it.

Related Literature

Our findings mainly relate to the literature on the anchoring of inflation expectations. There is a consensus that since the 1980s, and more so in the 2000s, inflation became anchored and monetary policy became more credible. The apparently relatively large effect that oil prices had on inflation and activity in the 1970s and early 1980s was followed by the ‘great moderation'. Leading macroeconomists sought to evaluate the contribution of monetary policy to the large impact of oil prices in the 1970s and even more so to the great moderation that ensued in the 1980s. Bernanke et al. (1997) argued that oil prices per-se did not have a large effect on the economy and that monetary policy response exacerbated their effect on the economy. Hooker (2002) did not rule out that the decline of the transmission between oil prices to the economy in the 1980s could have been due to effective monetary policy. Subsequent and influential research was more conclusive: Boivin and Giannoni (2006) found that by responding more strongly to inflation expectations, monetary policy has stabilized the economy more effectively in the post-1980 period. Blanchard and Gali (2010) and Blanchard and Riggi (2013) found that the improvement in the credibility of monetary policy explains a substantial part of the difference between the 2000s and the 1970s. Nakov and Pescatori (2010) found that around half of the reduced volatility of inflation is explained by better monetary policy alone, and that oil related effects explain around a third.

Our paper is related to the ongoing debate about the relevance and slope of the Phillips curve. While there is strong following for the view that “missing inflation” is consistent with the Phillips curve flattening or disappearing altogether [IMF (2013)], recent research has suggested otherwise. Blanchard (2016) shows that while it may have flattened, the Phillips curve is alive for the USA. Recent evidence from the Euro area, [Oinonen and Paloviita Riggi and Venditti (2015) and Bulligan and Viviano (2017)], is consistent with our findings and points to a steepening of the slope of the Phillips curve in the last decade. Borio and Filardo (2007) show that global economic slack has an increasingly important role in determining inflation since the early 1990's. Finally, Coibion and Gorodnichenko show that using expectations from survey data for the USA they can demonstrate the persistence of the Phillips curve which is not the case when market data on expectations is used. Moreover, they point to the high elasticity of survey expectations to oil prices. We can reconcile these separate findings using a global version of the Phillips curve, extracting a global aggregate demand factor from commodity prices and using principal component methods to remove country specific and idiosyncratic factors.

This paper also relates to the vast literature that studies the underlying forces in the market for crude oil [Kilian (2008b); Hamilton (2009a,b); Baumeister and Peersman (2013); Kilian and Murphy (2014); Fueki et al. (2018)]. Our approach is similar to the one taken by Kilian (2009) who uses a measure of global economic activity based on freight rates of dry cargo to identify the underlying shocks in the crude oil market. This measure, as well as ours, is designed to capture the main forces that drive the demand for a large group of commodities. Baumeister and Kilian (2016b) use this measure of global activity to examine the decline in oil prices in the second half of 2014. They find that the decline in prices was due to a momentum effect of positive supply shocks in earlier periods as well as unexpected adverse developments in global activity (they also highlight the role of storage demand driven by a shift in expectations). Our analysis of oil prices portrays a similar narrative.

Our decomposition of oil prices exploits the link between these prices and the prices of other commodities, using the first principal component of commodity prices. There is a large literature studying the linkage between prices of oil and other commodities [Baffes (2007); Du et al. (2011); Baumeister and Kilian (2014); Hassler and Sinn (2016); see Serra and Zilberman (2013) for a survey], and it points to several aspects of this linkage. First, prices of crude oil and other commodities are affected by global demand for the aggregate output. Second, crude oil enters the production function of other commodities through the use of various energy-intensive inputs. Third, some commodities can be used to produce substitutes to crude oil (e.g., corn and sugar for ethanol production), linking their demand to occurrences in the energy market. Finally, changes in the price of oil affect disposable income and thus influence the demand for other commodities. Note that out of the four mentioned links between prices of oil and other commodities, only the first two can explain contemporaneous co-movement of prices in a broad and diverse group of commodities such as we use. In accordance with previous studies [Barsky and Kilian (2001); Kilian (2009); Kilian and Murphy (2014)], we provide evidence that the global aggregate demand factor is more dominant in explaining the co-movement of prices, and that the pass-through from oil prices to other commodity prices is limited.

The rest of the paper is organized as follows. Section 2 specifies our methodology for testing the sources of change in oil prices and the anchoring of medium-term inflation expec­tations; In section 3 we estimate the effect of global demand on inflation expectations using a structural framework. Section 4 provides some robustness checks by examining household inflation expectations and excluding the period around the collapse of Lehman Brothers. Section 5 discusses possible implications for our findings and Section 6 concludes.

2 Accounting for the Change in the Relationship be­tween Oil Prices and Medium-Term Inflation Expec­tations

The increasing correlation between inflation expectations and oil prices following the onset of the global crisis may be suggestive of a structural change in the anchoring of medium term inflation expectations. Beechey et al. (2011) already noted that using oil price shocks has some advantages in comparing the anchoring of inflation expectations across countries since these are uniform shocks and since advanced economies have similar energy intensities. Extending Gurkaynak et al. (2005), they test for the anchoring of inflation expectations by regressing changes in far-ended inflation expectations on shocks to macroeconomic variables. If inflation is well anchored, these shocks, in particular oil price shocks, should not have a statistically significant effect on medium-term inflation expectations. In this section we employ Beechey’s et al. (2011) framework and test the anchoring of medium term inflation expectations. We extend their analysis in two ways: first, we differentiate between changes in oil prices induced by shifts in global aggregate demand and other sources of oil price changes. This refinement is important because, with the exception of the flexible CPI in­flation targeting rule [Svensson (2000)], monetary policy reacts differently to supply shocks and to demand shocks. While some degree of accommodation of supply shocks could be viewed as socially optimal [Rogoff (1985)] and could be incorporated in inflation expecta­tions for the medium-term, a perceived accommodation of demand shocks contrasts with inflation targeting and raises questions about the effectiveness or credibility of the monetary regime. Secondly, we extend the empirical investigation to include the period following the onset of the global financial crisis. In this period monetary policy was operating in hitherto uncharted territory of quantitative easing and negative interest rates. We present evidence that medium term inflation expectations became significantly more correlated with changes in oil prices induced by shocks to global output gap.

2.1 Estimating Global Aggregate Demand

To estimate the information on the global output gap embedded in oil prices we use the first principal component of commodity prices. While we refer to alternative methods of assigning demand and supply factors to oil price changes (see Section 1), we suggest that our measure is natural and transparent. We begin by describing the data and the methodology of principal component analysis. Subsequently, we analyze the relation of the estimated factor to global aggregate demand.

2.1.1 Data and Methodology

We use a panel of 20 commodity price indexes that were included in the S&P GSCI index in 2015. The data spans over the period 2000-2017 and includes commodities from five groups: agricultural commodities, livestock, industrial metals, precious metals, and energy (the commodities are listed in Table 1). In order to focus on fundamental co-movements of prices, we use monthly averages of commodity prices. One way to capture the co-movement of commodity prices is to take the first principle component of the levels of commodity prices (this factor accounts for 64 percent of the variation in commodity prices). However, we

Image pbz

 

convert the data to differenced logs of prices in order to avoid issues of non-stationarity. Finally, following the common practice in principal component analysis, all the series are standardized.

In our sample, the first principal component of rates of change in commodity prices, pc^cmd, explains 29 percent of the variance in the data. The loadings of all variables are positive (Table 1), so the first principal component captures the positive co-movement in all commodity prices. This justifies our interpretation of the first principal component as a proxy for global aggregate demand. In Section 4 we perform robustness checks that show that our measure is not driven by the fact that all commodities are quoted in USA Dollars and that the co-movement is not capturing the effect of oil prices on other commodities.

2.1.2 First Principal Component of Commodity Prices and Global Aggregate Demand

As we saw in Figure 2 the annual rates of change of the first principal component , pc^cmd (the unprocessed principal component based on monthly data is presented in Figure 11 in the Appendix) tracks very well the two measures of the global output gap. In the first measure, output is de-trended using an HP filter and in the second measure we use a trend of a ten year moving average growth rate (in the second method the level of the output gap is normalized to equal the HP filter gap in 2005). The correlation between the principal component of commodity prices and the change in output gap is 0.85 using either measure of the output gap.

2.2 Idiosyncratic Components of Oil Prices

We derived our measure of global output gap embedded in oil prices using first principal component of commodity prices. We can decompose the change in oil prices into two com­ponents:

Image egm

Where pc^cmd is the first principal component of commodity prices and the residual ut captures all remaining remaining variables affecting the change in oil prices. In this section we propose a more detailed specification of oil price decomposition that directly identifies some of the idiosyncratic forces that drive oil prices and show that they are largely orthogonal to our measure of global output gap. Showing that our measure of global demand is not correlated with oil specific factors strengthens our claim that the first principle component

Image ldh

of commodity prices can be used to identify the effect of global demand shocks on inflation expectations. We focus on two idiosyncratic components of oil prices, one from the supply side and the second from the demand side. From the supply side, we examine OPEC’s efforts to control prices of crude oil. These efforts may vary across time, depending on OPEC members’ objectives and their ability to collude to promote these objectives. From the demand side, we focus on the idiosyncratic demand shocks to oil driven by extreme weather conditions. The detailed breakdown of the change in oil prices to global demand, OPEC’s behavior and weather conditions is shown in Appendix C.

To estimate the effect of OPEC’s policies on crude oil prices, we assembled a novel data series that will serve as a proxy for the cartel’s operations. In each month of our sample period, we examine articles published in the London Times that refer to OPEC. We classify each article as either indicating supply expansion by OPEC, supply contraction or as neutral articles. Our proxy is then constructed as the net number of “expansionary articles”, meaning the number of articles classified as expansionary, minus the number of articles classified as contractionary (Figure 3). The sign of the proxy captures the objective of the cartel’s operations (negative indicating supply contraction, and positive indicating expansion), while the absolute size captures their magnitude.

For the measure of idiosyncratic demand shocks driven by extreme weather conditions, we examine global temperature data. The NCEI provides five monthly data sets, one for each continent, of seasonally adjusted temperature data. The rational of using this data is that weather conditions affect the usage of heating or cooling devices which are usually energy intensive, thus affecting the idiosyncratic demand for oil.

2.3 Estimating the Correlation between Oil Prices and Inflation Expectations

We now turn to examine the sources of the increase in the correlation between oil prices and inflation expectations, exploiting the decomposition of oil prices we performed in the previous section (Equation (1)). The increase in the correlation can be the result of two possible developments; First, it is possible that one of the factors that drives oil prices has become more correlated with inflation expectations. Alternatively, it may be that the magnitude of the correlation did not change but one of the factors became more dominant in determining oil prices in recent years. We claim that the elasticity effect dominates the composition effect. Specifically, we show that from the onset of the crisis, the correlation between the global aggregate demand factor embedded in oil prices and inflation expectations increased.

Since we are interested in the main factors that link oil prices and inflation expectations, we wish to ignore idiosyncratic components of breakeven inflation rates. We exploit the fact that our economies are anchors of the global monetary regime and pursue a similar inflation target. Therefore, we focus on the main factors that drive global inflation expectations. For this purpose we extract, p

To estimate the source of the correlation between inflation expectations and oil prices we regress Apc^, the change in the first principal component of five-year breakeven inflation rates, on decomposed oil prices allowing for a different correlation before and after the global financial crisis:

Image ele

Image yep

Where a1pc^cmd is our estimate for the global aggregate demand factor embedded in oil prices, Ut is the remaining component of oil prices (fitted value and residual from Equation, respectively) and dprecrit is a dummy for the pre-crisis period (2004M01-2008M08). Since the response of global aggregate demand to other shocks is not instantaneous , it is reasonable to assume that it is exogenous in Equation. However, we cannot account for all the determinants of the “remaining components” of oil prices and thus cannot assert that Ut is exogenous. We therefore use as instruments the variables derived in section 2.2: the proxy for OPEC’s cartelistic behavior and a factor of weather variables, both interacted with the pre-crisis period dummy. Two-stage least square estimation results, which are depicted in Figure 4, shed some light on the observed change in the correlation between changes to inflation expectations for the medium term and changes to oil prices.

Prior to the global crisis, changes in oil prices stemming from either global aggregate demand or other components, had a small and similar effect on inflation expectations. This suggests that even if the composition of factors that drive oil prices has changed, it cannot by

Image doz

 

itself explain the increase in correlation between oil prices and inflation expectations in recent years. In the post-crisis period the picture is very different. We cannot reject the hypothesis that the effect of the remaining components of oil prices on expectations remained stable. However, the effect of global aggregate demand increased significantly. Since we breakdown our sample to pre-crisis and post-crisis according to the timing of the collapse of Lehman Brothers in September 2008, one might argue that the correlations we document are driven by the sharp drops of both inflation expectations and commodity prices in the months that followed the collapse. Though there is no a-priori reason to partition the sample differently, for robustness we estimate Equation (2) controlling for the crisis years 2008-2009. In this exercise we also see a substantial increase in the coefficient of the global aggregate demand factor since the crisis, but not in that of the idiosyncratic component. See section 4.2 for a detailed discussion.

Our results suggest that the information embedded in oil prices regarding global activity has become much more dominant in the formation of inflation expectations, even at the five-year horizon. The contribution of the global aggregate demand factor to the correlation of global inflation expectations over time is depicted in Figure 5 and it seems that in the post-crisis period it accounts for a substantial part of the development in expectations. In the following section we offer an explanation for this change using a rational expectations framework.

3. A Rational Expectations View on the Anchoring of Inflation Expectations

We showed that global medium-term inflation expectations became more correlated with oil prices since the global financial crisis, and this is due to a stronger effect of global aggregate demand on inflation expectations. A possible concern for policymakers is that the increased sensitivity of breakeven inflation rates to oil prices may indicate an erosion in the anchoring of expectations and the credibility of the inflation targeting regime, especially since mone­tary policy has been operating in formerly uncharted territories of quantitative easing and negative interest rates [e.g., IMF (2016)].

To address the question whether inflation expectations became un-anchored, we turn to examine the increased correlation between oil prices and five-year breakeven inflation rates in a context of a structural model with rational expectations. The structural framework, together with our decomposition of oil prices, allows us to identify channels through which inflation expectations react more strongly to the global demand factor embedded in oil prices. We are able to show that this change can be attributed to the increase in the slope of the Phillips curve. Furthermore, we find no evidence that expectations became more adaptive since the criss, meaning that they have so far remained forward-looking. These results hold for market expectations for a five-year horizon as well as for household expectations for a twelve-month horizon. We conclude that the increased responsiveness of inflation expectations to the global aggregate demand factor embedded in oil prices does not indicate un-anchoring of expectations as formulated in Beechey et al. (2011) or Blanchard (2016).At the same time we find that although the coefficient of idiosyncratic oil price changes on breakeven inflation rates increased, the increase is not statistically significant. Moreover, unlike Coibion and Gorodnichenko (2015), we do not find, controlling for the information on global aggregate demand embedded in oil prices, that their effect on household survey-based expectations increased; it actually (insignificantly) declined.

3.1 Rational Expectations with Optimal Policy Rule

To examine the change in the correlation between changes in medium term global inflation expectations and changes to global output gap and other determinants of oil prices, we introduce a framework of how expectations are formed. We consider the semi-structural model of Orphanides and Williams (2004) which was also used in Beechey et al. (2011) to examine the anchoring of inflation expectations. The model consists of a Phillips curve and an IS curve as follows:

Image bkn

Where nt is the annual rate of inflation at time t, nt+1/t is the one-period ahead expected inflation, yt is the output gap, rt is the real interest rate, r* is the long-run real rate, et is a cost-push shock and ut is a demand shock. The model is closed with the following policy rule that minimizes a weighted average of the variances of the output gap and of deviation

In this model, rational expectations for inflation take the following form:

Image dpb

Since we wish to employ our decomposition of oil prices that identifies changes in the output gap (see Section 2.1.2) we convert this equation to a difference equation:

Image rfb

In accordance with this specification, we estimate the following regression model, using our proposed decomposition of oil prices to account for changes in the output gap and cost- push shocks and allowing for a structural change after the global financial crisis:

Image non

Where pc^7- and pc^ are the first principle components of five-year inflation expectations and annual inflation, respectively, rescaled to match the data. aipc^cmd is our estimate for the global aggregate demand factor embedded in oil prices and Ut is the remaining component of oil prices (fitted value and residual from Equation (1), respectively).

As explained in Section 2.3, we assume that the response of global aggregate demand to other shocks is not instantaneous and therefore assume pc^cmd is exogenous in Equation (8). Since we cannot assert that Ut is exogenous, we use the instruments derived in Section to capture exogenous changes in oil prices. Two-stage least square estimation results are depicted in Panel A in Figure 6 (detailed results appear in Table 3 in the Appendix).

Our results indicate that the only coefficient that significantly changed after the crisis is that of global aggregate demand. Other than that variable, no other determinant of inflation expectations has a different effect since the onset of the crisis. Specifically, the adaptive component of expectations is low and stable in our sample. There was also no significant change in the response of expectations to changes in oil prices unrelated to global aggregate demand. Examining the structural specification of inflation expectations (7) we find that an increase in the coefficient on the output gap with stability of the other coefficients is necessarily due to an increase in a, the perceived slope of the Phillips curve, while the adaptive coefficient 1 - S remained stable. The analysis we perform in the following section provides additional evidence for this conclusion.

Image ffm

The fact that expectations did not become more adaptive or more responsive to oil- specific price changes since the crisis implies that they have so far remained anchored and that the concerns for the credibility of inflation targeting were premature. Instead, our results point to a perceived structural change in the Philips curve that made inflation expectations more responsive to global aggregate demand and thus more correlated with oil prices which contain information regarding global output.

While we do not directly estimate the Phillips curve and therefore cannot take a stand on whether the slope has indeed changed or was it merely perceived by the public to have changed, some recent papers find evidence of structural change in the Phillips curve in recent years. Riggi and Venditti (2015) show an increase in the sensitivity of inflation to the output gap in the Euro area since the sovereign debt crisis and offer two structural explanations for this change. First, lower nominal rigidities, and second, a fall in the number of firms in the economy that lowers the elasticity of demand. For the USA economy, Stella and Stock (2012) find evidence of a stronger inflation-unemployment relationship since the global financial crisis. We believe that a fruitful direction for future research will be to use our measure for the global output gap, namely the first principal component of commodity prices, to test for changes in the Phillips curve in a global perspective.

3.2 Rational Expectations without Specifying the Policy Rule

Monetary policy has operated since the global financial crisis in an environment of interest rates approaching the ’effective lower bound’ and saw an extended use of unconventional policies. We, therefore, consider an alternative specification of rational inflation expectations which is agnostic to the monetary policy rule. In Section 3.1 we used the Phillips curve and the IS curve (Equations (3) and (4), respectively) as well as an optimal monetary rate rule (5) to formulate rational inflation expectations. However, using only (3) and (4), we can construct an alternative formulation of expectations which does not assume any structure of the monetary policy rate:

Image cgr

We estimate the following model which adds to (8) a measure of the change in the global monetary rate, pc^1 - the first principal component of the change in the monetary interest rate in the USA, UK and the Euro area:

Image nee

Following our discussion in the previous section, it is reasonable to assume that global demand is exogenous in (11). However, the monetary interest rate is clearly endogeneous and we also treat Ut as such. We thus estimate (11) using the lag of pcf-1 as an instrument for the monetary rate, and the instruments detailed in Section 2.3 for Ut.

Two-stage least-square estimation results of model (11) are depicted in Panel B of Figure 6 and they portray a similar picture to the one portrayed by the estimation of Equation (8), namely that the effect of global aggregate demand on medium-term inflation expectations in­creased substantially since the global financial crisis while the effect of the other components remained stable. Examining the structural specification (10) confirms that these results may only be explained by a perceived rise in the slope of the Phillips curve, a, while the parameter of inflation adaptiveness, 1 - 0, remains unchanged. We conclude that there was no change in the anchoring of inflation expectations after the crisis, but rather a perceived structural change that made inflation expectations more sensitive to global aggregate demand.

4. Robustness

In this section we perform four main robustness checks to our results. The first is to use household expectations instead of expectations derived from financial markets. The second is to show that our results are not driven by the financial turmoil around the collapse of Lehman Brothers. In the third and final sections we consider two possible reservations regarding our interpretation of the first principle component of commodity prices as reflecting changes in global aggregate demand: The direct effect of oil prices on other commodities and the effect of the USA Dollar exchange rate. Additional robustness checks are deferred to the appendix.

4.1 Household Inflation Expectations

So far, our analysis of inflation expectations focused on five-year breakeven inflation rates. While this measure is closely monitored by policymakers and is readily available for a sub­stantial set of countries, it has some shortfalls [Coibion and Gorodnichenko (2015)]. First, it is affected by financial factors such as risk and liquidity premiums. Second, one might argue that a five-year horizon is too long to consider in a context of a Phillips curve which is constructed to capture the effect of nominal rigidities. Therefore, in a way of robustness test, we repeat the analysis preformed in the two previous sections using household surveys, a measure of one-year inflation expectations that is commonly used in the literature examining the Phillips curve.

Data availability only allows us to use quarterly data, however, we are able to extend our sample to 2000Q1-2017Q3. We extract the first principal component of household surveys for the USA, UK and Euro area, pcstur. Following the convention in the Phillips curve literature and more recently Coibion and Gorodnichenko (2015), we estimate the two models of rational expectations in levels (Equations (6) and (9)) . For this purpose we first decompose the log level of oil prices

Image pbb

where pcltmd is the first principal component of levels of commodity prices (in log terms). This factor serves as an estimate of the global output gap so we interpret the fitted value of the regression, Y'ipctmd, as the component of oil prices driven by the level of global aggregate activity. The change in the residual, Avt, is a proxy for cost-push shocks. Using these measures we estimate the following two models of rational inflation expectations, one without the monetary rate and one with the monetary rate:

Image xcx

2SLS estimation results of these models are depicted in Figure 7 (detailed results are reported in Table 4 in the Appendix). Both specifications indicate, as found in Coibion and Gorodnichenko (2015), that the effect of oil prices on consumers’ expectations is significant and has increased. However, the increase can be attributed to the effect of global aggregate demand embedded in oil prices, while the effect of all other variables remained stable. This means that similarly to breakeven inflation rates, household expectations have not become more adaptive since the crisis but are consistent with a perceived increase in the slope of the Phillips curve.

Coibion and Gorodnichenko (2015) find that incorporating household expectations in the estimation of a standard Phillips curve for the USA resolves the puzzle of the “missing disinflation” in the aftermath of the global financial crisis. This is due to the rise in household expectations at that period which they attribute to a rise in oil prices. Our results can be interpreted as indicating that it is not the rise in oil prices themselves that affected household expectations at the aftermath of the crisis, but rather the information regarding the global output gap that is embedded in them.

4.2 Excluding the Period of the Financial Turmoil Around the Collapse of Lehman Brothers

In Section 2.3 we showed that since September 2008 global aggregate demand conditions embedded in oil prices have a stronger effect on medium-term inflation expectations. One might argue that the strong effect stems from a short period following the collapse of Lehman Brothers and does not reflect the later period. Nevertheless, in this section we show that while the months following Lehman’s collapse contributed to our identification, they do not fully account for our main results. Namely, we find that even if we disregard a wide period around Lehman’s collapse, the effect of global aggregate demand conditions on inflation expectations increased in the post-crisis period relative to the pre-crisis period. On the other hand, the effect of the remaining component of oil remained unchanged.

We consider a model which is based on Equation (2) but instead of a single breakpoint in the sample, we partition the sample to three periods: 2003M03-2007M12 (pre-crisis), 2008M01-2009M12 and 2010M01-2017M08 (post-crisis).

Image i94

Image ucv

 

Where d03-07 and d08-09 are dummy variables for 2003M03-2007M12 and 2008M01- 2009M12, respectively. We can then compare the coefficient of pre-2007 to those of post 2010, disregarding an extensive period around the collapse of Lehman Brothers. The results of this exercise are compared to our baseline results in Table 5. In both models the coef­ficient of the global aggregate demand factor increases significantly after the crisis, while the coefficient of the remaining components is stable (the difference between the periods is insignificant).

Next we repeat the same exercise on our structural equations and estimate variants of Equations (8) and (11) that include interactions with the dummies d03-07 and d08-09 instead of dprecrisis. Our main results, shown in figure 8, are that the impact of changes in global output gap on changes in breakeven expectations increased significantly, whereas the effect of idiosyncratic changes to oil prices did not significantly change. Admittedly, the significance of the difference between the two periods is lower than in our benchmark regressions, however, it is significant at the 5 percent level.

Image joi

4.3 The direct effect of oil prices on the other commodities

We consider the possibility that the co-movement in commodity prices captured by pc^cmd may be related to energy prices since manufacturing of all commodities requires some use of energy. If this effect is significant, pc^cmd may be capturing the evolution of energy prices rather than global aggregate demand. However, we argue that energy prices have only a modest effect on other commodity prices, so they do not dominate the first principal component.

First, we note that the energy component contained in agriculture and metal industries is small. To illustrate this point, we examine data from the USA Department of Commerce regarding six industries that best match the S&P non-energy commodities. In each of these six industries we calculate the value of energy-intensive inputs, as a share of total output in that industry. As specified in Table 2, the share of total output that can be associated with energy-intensive inputs is lower than 17 percent in all six industries. This is consistent with

Image xxv

 

the findings of Baffes (2007) which reports pass-through rates of 0.11-0.19 from oil prices to prices of metals and agricultural commodities.

Second, we perform a Granger Causality test between pc^cmd and the monthly rate of change in the S&P energy index. The test indicates that we cannot reject the hypothesis that the energy index does not Granger cause pc^cmd (F-statistic of 0.70). This means that given past information regarding the first principal component, energy prices have no significant contribution to forecasting pd^cmd. The result supports our argument that energy prices have a limited effect on pd^cmd.

Interestingly, a Granger Causality test for the other direction shows that pc^cmd Granger causes the monthly rate of change in the energy index (F-statistic of 4.51 for the null hypoth­esis that pc^cmd does not Granger cause the monthly rate of change in the energy index).

This result suggests that energy prices are highly influenced by global aggregate demand. An even stronger result is obtained when we test the hypothesis that pc^cmd does not Granger cause the monthly rate of change in the prices of Brent crude oil (F-statistics of 4.89).

4.4 Controlling for the USA Dollar exchange rate

Since all commodities are denominated in USA Dollars it may be argued that our measure of global demand extracted from the first principle component of all commodities might reflect changes in the USA Dollar exchange rate. While it is reasonable to assume that changes in the USA Dollar exchange rate are endogenous to shocks to global demand we nevertheless control for the exchange rate in our baseline regression. In this robustness we also subject the data to the previous robustness check and exclude the period around the Lehman Brothers’ crisis.

We regress the following two stage regression:

Image ddt

Where aipc^cmd is our measure for global demand and the independent variable is the change in the log of the Dollar trade weighted exchange rate (DXY). We take the residuals Vt from equation (15) above as an adjusted measure of global aggregate demand and re­estimate Equation (14) and the two structural equations, controlling the years 2008-2009 (as in Section 4.2). The results of equation (14) are reported in table 5 in the appendix and the results of the structural equations are depicted in figure 9.

Controlling for the exchange rate, we still find a significant increase in the effect of global aggregate demand on global inflation expectations after the crisis (the results are actually slightly more robust with respect to the estimation in the preceding exercise of controlling for the crisis years). However, this model also shows a marginally significant increase in the effect of the idiosyncratic component of oil prices.

5 Discussion: Implications of the Increased Correla­tion Between Oil Prices and Inflation Expectations

Our results show that inflation expectations for the medium-term are affected by oil prices and that this effect increased since the GFC. Decomposing oil prices, we showed that their reaction to oil specific shocks is small and remained stable, consistent with the accepted practice to look through supply shocks [Rogoff (1985); Ireland (2007)]. However, our results also show that inflation expectations react more to global aggregate demand shocks than previously. Since central bankers view the stability of medium and long term inflation expectations - anchoring - as indicators of the credibility of inflation targeting, our result could indicate an erosion of that credibility. In particular, the results may suggest that the

Image rua

public perceives that monetary authorities either look through demand shocks or may have changed their policy rules. These perceptions could lead to un-anchoring of expectations.

Embedded in a structural New Keynesian model with rational expectations, the same results suggest that, so far, expectations have not become un-anchored. The increasing correlation between inflation expectations and demand shocks is consistent with a structural change in the slope of the Phillips curve. Under an optimal policy rule, an increase in the slope of the Phillips curve (even if it is only a perceived rise), given a Taylor type monetary rule, should call for a weaker response of the monetary rate to deviations of inflation from target. In terms of the model presented in Section 3.1, this means that policymakers should reduce 0.

It is useful to clarify what exactly is meant by un-anchoring. In the context of a reduced form New Keynesian model (Phillips curve) un-anchoring could be related to changes in the degree of backward looking behavior or persistence of past expectations. Our finding for the global Phillips curve, similar to those of Blanchard (2016) for the USA economy show no such change (for financial market data and household survey data). Another approach is to test whether expectations react more to idiosyncratic shocks. Our results indicate, in the spirit of Beechey et al. (2011) that we cannot reject the hypothesis that this did not happen in the case of global five year breakeven rates. Our findings, however, are consistent with a steepening of the Phillips curve which means that for a given demand shock, inflation volatility increases. Increasing inflation volatility also means that rational agents’ inflation expectations for the short and medium term become less stable.

The implications of our results are that care should be taken by monetary authorities in interpreting the correlation between oil prices and inflation expectations. In particular, the assessment by policy makers [IMF (2016)] of un-anchoring may lead to policy actions, including forward guidance, that may actually contribute to un-anchoring of longer term forward inflation expectations. While this is beyond the scope of this paper, beginning in 2014 we witness a sharp decline five-year to five-year forward inflation expectations. What could explain this change? We offer two possible, mutually non-exclusive, explanations. The first is that monetary policy had been operating since 2008 in a new environment where interest rates have reached the “effective lower bound”. At the same time unconventional monetary tools such as quantitative easing and forward guidance are employed. Our findings suggest that the public may perceive these measures as less effective in restoring inflation to its target. Moreover, according to Morris and Shin (2018), this could have been amplified by non-conventional monetary tools. Central bank communications that tied future policy to long term forward expectations could have led to a vicious circle of declining long term expectations.

The second explanation relates to a perceived change in monetary policy objectives. Before the global financial crisis monetary authorities followed, or were expected to follow, a Taylor rule that puts a large weight on meeting the inflation target and a lower weight on stabilizing output. At that period inflation expectations were firmly anchored at the two percent level. However, the financial crisis of 2008 stressed the importance of financial stability in maintaining output growth and price stability. Consequently, several central banks adopted “leaning against the wind” approaches in recent years [ECB (2010); Svensson (2014); Filardo and Rungcharoenkitkul (2016)]. While the effectiveness of using the central bank rate to achieve macro-prudential goals, and particularly to contain asset price bubbles, has been under debate [Gall (2014); Gall and Gambetti (2015)], and is considered by some to be a blunt tool in dealing with financial stability issues [Bernanke (2010); Blanchard et al. (2010)], nevertheless, monetary authorities became more attentive to financial conditions in recent years. It could be that the public interpreted this as a decline in the commitment to uphold the inflation target in the medium term. A variant of this explanation is that when inflation deviates below the target, the public believes that monetary authorities will be less aggressive in attempting to move it back into the target zone, i.e., that the weight on inflation in the Taylor rule is asymmetric with respect to positive and negative deviations from the target.

6. Conclusions

We used the first principal component of a variety of commodity prices to decompose the changes in oil prices to those emanating from a global demand factor and those that arise due to oil specific ones. We use this decomposition to analyze the increase in the correlation between oil prices and inflation expectations following the onset of the global financial crisis and find that it is mainly due to a stronger effect of global aggregate demand embedded in oil prices on expected inflation.

We compute global inflation and monetary regime variables using principal component method and estimate a global Phillips curve. We cannot reject the hypothesis that expec­tations remained anchored. Instead, our structural estimation suggests that the increased volatility of inflation expectations with respect to global demand factors can be explained by an increase in the perceived slope of the Phillips curve. Our findings have implications for monetary policy.

The high degree of covariation in major global macroeconomic aggregates allows to ex­tract global factors using principle component analysis. While we used this methodology to focus on the global factors themselves, policy makers in advanced small open economies can use it to identify domestic factors that they can hope to control and hone their policies accordingly. Specifically, the principal component we use as a proxy for global aggregate de­mand can be readily employed to monitor in real time global aggregate demand conditions. Moreover, this factor can be useful for macroeconomic empirical research that uses higher temporal frequency data, either as a proxy or as an instrument. For example, the proxy can be used to infer the contribution of global aggregate demand shocks on monthly, country specific, price level data. Another example is to use the variable as an instrument in research that uses the CPI which is usually determined simultaneously with the left hand variable in question. Finally, one can use our proxy to revisit some of the studies on monetary policy and oil prices since the 1970s.

References

Alquist, R. and O. Coibion (2014). Commodity-price comovement and global economic activity. NBER Working Paper (20003).

Baffes, J. (2007). Oil spills on other commodities. Resources Policy 32(3), 126-134.

Barsky, R. B. and L. Kilian (2001). Do we really know that oil caused the great stagflation? a monetary alternative. NBER Macroeconomics annual 16, 137-183.

Barsky, R. B. and L. Kilian (2004). Oil and the macroeconomy since the 1970s. The Journal of Economic Perspectives 18(4), 115-134.

Baumeister, C. and L. Kilian (2014). Do oil price increases cause higher food prices? Economic Policy 29 (80), 691-747.

Baumeister, C. and L. Kilian (2016a). Lower oil prices and the u.s. economy: Is this time different? Brookings Papers on Economic Activity.

Baumeister, C. and L. Kilian (2016b). Understanding the decline in the price of oil since june 2014. Journal of the Association of Environmental and Resource Economists 3(1), 131-158.

Baumeister, C. and G. Peersman (2013). The role of time-varying price elasticities in accounting for volatility changes in the crude oil market. Journal of Applied Econometrics 28(7), 1087-1109.

Beechey, M. J., B. K. Johannsen, and A. T. Levin (2011). Are long-run inflation expectations anchored more firmly in the euro area than in the united states? American Economic Journal: Macroeconomics 3(2), 104-129.

Bernanke, B. S. (2010). Monetary policy and the housing bubble. In speech at the annual meeting of the American Economic Association, Atlanta, Georgia, Volume 3.

Bernanke, B. S., M. Gertler, M. Watson, C. A. Sims, and B. M. Friedman (1997). Systematic mon­etary policy and the effects of oil price shocks. Brookings papers on economic activity 1997(1), 91-157.

Blanchard, O. (2016). The u.s. phillips cvurve: Back to the 60s? Peterson Institute for International Economics Policy Breif (PB16-1), 1-4.

Blanchard, O., G. DellAriccia, and P. Mauro (2010). Rethinking macroeconomic policy. Journal of Money, Credit and Banking 42(s1), 199-215.

Blanchard, O. J. and J. Gali (2010). The macroeconomic effects of oil shocks: Why are the 2000s so different from the 1970s? In J. Gall and M. Gertler (Eds.), International Dimensions of Monetary Policy, pp. 373-428. University of Chicago Press.

Blanchard, O. J. and M. Riggi (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., L. Guerrieri, and L. Kilian (2012). Monetary policy responses to oil price fluctua­tions. IMF Economic Review 60(4), 470-504.

Boivin, J. and M. P. Giannoni (2006). Has monetary policy become more effective? The Review of Economics and Statistics 88(3), 445-462.

Borio, C. and A. Filardo (2007). Globalisation and inflation: New cross-country evidence on the global determinants of domestic inflation. BIS Working Paper (227).

Bulligan, G. and E. Viviano (2017). Has the wage phillips curve changed in the euro area? IZA Journal of Labor Policy 6(9).

Byrne, J. P., G. Fazio, and N. Fiess (2013). Primary commodity prices: Co-movements, common factors and fundamentals. Journal of Development Economics 101, 16-26.

Chen, D. C., X. Gong, E. Raju Huidrom, J. Z. Vashakmadze, and T. Zhao (2015). Understanding the plunge in oil prices: Sources and implications. The World Bank: Washington, DC, USA.

Ciccareli, M. and B. Mojon (2010). Global inflation. The Review of Economics and Statistics 92 (3), 524-525.

Coibion, O. and Y. Gorodnichenko (2015). Is the phillips curve alive and well after all? inflation expectations and the missing disinflation. American Economic Journal: Macroeconomics 7(1), 197-232.

Diebold, F. X., L. Canlin, and V. Z. Yue (2008). Global yield curves dynamics and interactions: a dynamic nelson-siegal approach. Journal of Econometrics 146(2), 351-363.

Du, X., L. Y. Cindy, and D. J. Hayes (2011). Speculation and volatility spillover in the crude oil and agricultural commodity markets: A bayesian analysis. Energy Economics 33(3), 497-503.

European Central Bank (2010). Asset price bubbles and monetary policy revisited. Monthly Bulletin November, 71-83.

Filardo, A., M. J. Lombardi, et al. (2014). Has asian emerging market monetary policy been too procyclical when responding to swings in commodity prices? BIS Papers chapters 77, 129-153.

Filardo, A. J. and P. Rungcharoenkitkul (2016). A quantitative case for leaning against the wind. BIS Working Paper (594).

Fueki, T., H. Higashi, N. Higashio, J. Nakajima, S. Ohyama, Y. Tamanyu, et al. (2018). Identifying oil price shocks and their consequences: the role of expectations in the crude oil market. BIS Working Paper (725).

Gall, J. (2014). Monetary policy and rational asset price bubb. The American Economic Re­view 104(3), 721-752.

Gall, J. and L. Gambetti (2015). The effects of monetary policy on stock market bubbles: Some evidence. American Economic Journal: Macroeconomics 7(1), 233-257.

Gertler, M. and P. Karadi (2015). Monetary policy surprises, credit costs, and economic activity. American Economic Journal: Macroeconomics 7(1), 44-76.

Gurkaynak, R. S., A. Levin, and E. Swanson (2010). Does inflation targeting anchor long-run inflation expectations? evidence from the us, uk, and sweden4. Journal of the European Economic Association 8(6), 1208-1242.

Gurkaynak, R. S., B. Sack, and E. Swanson (2005). The sensitivity of long-term interest rates to economic news: evidence and implications for macroeconomic models. The American Economic Review 95(1), 425-436.

Hamilton, J. D. (2009a). Causes and consequences of the oil shock of 2007-08. Brookings Papers on Economic Activity.

Hamilton, J. D. (2009b). Understanding crude oil prices. Energy Journal 30(2), 179-206.

Hassler, J. and H.-W. Sinn (2016). The fossil episode. Journal of Monetary Economics 83, 14-26.

Hooker, M. A. (2002). Are oil shocks inflationary?: Asymmetric and nonlinear specifications versus changes in regime. Journal of Money, Credit, and Banking 34 (2), 540-561.

IMF (2013). The dog that did not did not bark: Has inflation been muzzled or was it just sleeping? World Economic Outlook April.

IMF (2016). Global disinflation in an era of constrained monetary policy. World Economic Outlook (WEO) October(Ch. 3), 121-170.

Ireland, P. N. (2007). Changes in the federal reserve’s inflation target: causes and consequences. Journal of Money, credit and Banking 39(8), 1851-1882.

Kiley, M. T. (2014). The aggregate demand effects of short-and long-term interest rates. Interna­tional Journal of Central Banking.

Kilian, L. (2008a). The economic effects of energy price shocks. Journal of Economic Litera­ture 46(4), 871-909.

Kilian, L. (2008b). Exogenous oil supply shocks: how big are they and how much do they matter for the us economy? The Review of Economics and Statistics 90 (2), 216-240.

Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply shocks in the crude oil market. American Economic Review 99(3), 1053-1069.

Kilian, L. and B. Hicks (2013). Did unexpectedly strong economic growth cause the oil price shock of 2003-2008? Journal of Forecasting 32(5), 385-394.

Kilian, L. and D. P. Murphy (2012). Why agnostic sign restrictions are not enough: understanding the dynamics of oil market var models. Journal of the European Economic Association 10(5), 1166-1188.

Kilian, L. and D. P. Murphy (2014). The role of inventories and speculative trading in the global market for crude oil. Journal of Applied Econometrics 29 (3), 454-478.

Larkin, J. (2014). Examining the sensitivity of inflation to the output gap across euro area member states. Central Bank of Ireland Quarterly Bulletin, Q2, 50-51.

Mankiw, N. G. and R. Reis (2003). What measure of inflation should a central bank target? Journal of the European Economic Association 1 (5), 1058-1086.

Morris, S. and H.-S. Shin (2018). Central bank forward guidance and the signal value of market prices. BIS Working Paper (692).

Nakov, A. and A. Pescatori (2010). Oil and the great moderation. The Economic Journal 120(543), 131-156.

Oinonen, S. and M. Paloviita (2014). Updating the euro area phillips curve: the slope has increased. Technical report, Bank of Finland.

O’Neill, T., J. Penm, and R. Terrell (2008). The role of higher oil prices: A case of major developed countries. Research in Finance 24 (0), 287-299.

Orphanides, A. and J. Williams (2004). Imperfect knowledge, inflation expectations, and monetary policy. In the inflation-targeting debate, pp. 201-246. University of Chicago Press.

Perez-Segura, A. and R. J. Vigfusson (2016). The relationship between oil prices and inflation compensation. IFDP Notes. Washington: Board of Governors of the Federal Reserve System.

Reis, R. and W. Watson, Mark (2010). Relative goods’ prices, pure inflation and the phillips correlation. American Economic Journal: Macroecoonmics 2(3), 128-157.

Riggi, M. and F. Venditti (2015). Failing to forecast low inflation and phillips curve instability: A euro-area perspective. International Finance 18 (1), 47-68.

Rogoff, K. (1985). The optimal degree of commitment to an intermediate monetary target. The quarterly journal of economics, 1169-1189.

Serra, T. and D. Zilberman (2013). Biofuel-related price transmission literature: A review. Energy Economics 37, 141-151.

Stella, A. and J. H. Stock (2012). A state-dependent model for inflation forecasting.

Stock, J. H. and M. W. Watson (2011). Dynamic factor models. Oxford Handbook of Economic Forecasting 1, 35-59.

Svensson, L. (2014). Inflation targeting and” leaning against the wind”. International Journal of Central Banking 10 (2), 103-114.

Svensson, L. E. (2000). Open-economy inflation targeting. Journal of international eco­nomics 50(1), 155-183.

Tawadros, G. (2013). The cyclicality of the demand for crude oil: evidence from the oecd. Journal of Economic Studies 40(6), 704-719.

 

Image ubb

 

Appendix

A Oil Prices and Inflation Expectations

In this section we show more rigorously the observed relationship between oil prices and expected inflation (Figure 1). We estimate country-specific regressions of five-year breakeven inflation rates, beiri,t, on oil prices, allowing for a different effect before and after the global crisis. For each country i we estimate the following regression:

Image hnb

Where oilj,,t is the log price of Brent crude oil in domestic currency34, dprecrit is a dummy for the pre-crisis period (2004M01-2008M08), dlehi,t is a dummy variable that equals one circa September 2008 (indicating known liquidity problems in country i’s government bonds market).

Figure 10 depicts the estimated correlation between oil prices and five-year breakeven inflation rates (detailed estimation results are reported in Table 6 ). Similarly to the corre­lations reported in Figure 1, the regression results show a strengthening correlation between oil prices and medium-term inflation expectations after the onset of the global crisis.

 

B Methodology for Constructing the First Principal Component

In Section 2.1 we presented our estimator of global aggregate demand - the first principal component of commodity prices. We now briefly discuss the methodology for constructing this factor (see Stock and Watson (2011) for more details).

The first principal component is a factor that best explains the total variation in the data. For a data set of N variables over T periods, let Xt E RN denote the column vector of variables in period t E {1,...,T}. In our case there are N = 20 commodities and Xt is the vector of monthly changes in prices. The first principal component is the factor (/1,...,/T) E RT that, together with a loading vector A E RN, solves the least square problem,

Image bgg

The factor that solves this problem is a linear combination of the variables constructed as follows. Denote the sample variance matrix by E = T-1 y)^—1 XtX[ and let A be the normal­ized eigenvector of EE associated with the largest eigenvalue. The first principal component estimator is then given by /t = A'Xt and the loading vector is TV.

The first principal component of the monthly rate of change of the 20 commodity prices is depicted in Figure 11.

C Decomposing Oil Prices

Recall from equation (1) that the basic decomposition of oil prices is Aoilt = a0 + a1pe^cmd + ut, where pe^cmd is the first principal component of commodity prices and the residual ut captures all remaining remaining (orthogonal) variables affecting the change in oil prices. In the detailed decomposition (section 2.2) we added the proxy for OPEC’s operation and weather variables:

Image xes

Where opeere/t is the OPEC “net references” proxy, wt is a vector of the temperatures measured in the five continents, rs, s = 1,.., 4 are vectors of coefficients, and ds,t is a dummy variable for the season of the year.

Image jnu

Image cge

 

The least square estimator of the coefficient of pc^cmd in Equation (17) is 0.022 (s.e. 0.0018), compared to an estimator of 0.023 (s.e. 0.018) in Equation (1).

Figure 12 depicts the contribution of all elements in Equation (17) to the annual rate of change in oil prices. We see that the proxy for OPEC’s operation explains a substantial portion of the price changes in several periods (admittedly, the weather component has less explanatory power). For example, we see that expansionary operations of OPEC since mid- 2014 contributed considerably to the decline in prices. This is in line with our previous knowledge regarding the decreased ability of OPEC to collude in that period.

D Alternative Specifications

D.1 Basic Oil Price Decomposition

In this section we explore alternative specification for the decomposition of oil prices. Recall that the baseline specification, as presented in Equation (1), is:

Image cin

The OLS estimation results of the baseline model and several other specifications are summarized in Table 7. We examine different lag structures of the equation (columns (2)- (3)) and the use of the first principal component of non-energy commodities instead of pc^cmd (columns (4)-(5)). We find that the estimate for the coefficient of the first principal component is robust at around 0.02.

In column (6) of Table 7 we present estimation results for real prices. We repeat the procedure specified in section 2.1 with all prices divided by USA core CPI. This means that we extract the first principal component of real commodity prices, pc^rcmd, and use it to decompose real oil prices. In the final column of Table 7 we perform another robustness check and use the real prices of oil prices and a principal component of real prices of commodities excluding energy and gold, pc^rnen9, we also control for financial uncertainty by including the log of the VIX.

The final step in our analysis requires using the decomposition of oil prices to estimate the effect of global aggregate demand and other changes in oil prices on global expected inflation (Equation (2)). With either one of the decompositions specified in Table 7, the estimated coefficients in the final step are not significantly different from the ones in our baseline model, so we waive the presentation of the detailed results.

D.2 Detailed Oil Price Decomposition

This section specifies estimation results of the detailed decomposition of oil prices and the contribution of idiosyncratic components (Section 2.2). The first two columns of Table 8  present estimation results of the basic model (Equation (1)) and the detailed model (Equation (17)) of oil price decomposition. Comparing the two columns, we see that the coefficient of pCAcmd is estimated at around 0.02 in both models, indicating that the variables added in the second model are orthogonal to pc‘Acmd.

For a robustness check, we present an alternative model in the third column of Table 8. It is similar to the detailed specification from the second column, except for a different specification of the weather variables. Recall that in Equation (17) we use the temperature data from five continents, interacted with dummy variables for the seasons of the year. In the third column of Table 8, we estimate a model with dummy variables for calendar months:

Image pft

Where Wt is the vector of the temperatures in the five continents, Am, m = 1,..., 12 are vectors of coefficients, and dm,t is a dummy variable for the calendar month m. As seen in Table 8, the estimated coefficients of pc£Acmd and opecreft are essentially the same as those estimated in columns 1-2.

A leading topic in the public discussion regarding the 2014 oil price decline was tech­nological developments in the production of shale oil. As shale oil is a substitute for crude oil, technology developments in its manufacturing are expected to lower prices of crude oil. To test this effect, we examined references of shale oil in the London Times (Figure 13). There are not much references of shale oil prior to 2009 (45 references in the period 2000M01-2008M12, relative to 1317 in 2009M1-2017M08), and since 2014 the series of shale oil references, shalereft, is correlated with opecreft (partially by construction since some articles mention both OPEC and shale oil). Thus it is not surprising that shale oil references do not contribute to the estimation of oil price changes (forth column of Table 8).

D.3 Alternative Data Frequencies

In this section, we test the sensitivity of our results to data frequency. In the baseline estimation we used monthly averages of daily data. This frequency conversion was used for the estimation of the first principal component, the decomposition of oil prices, and the analysis of breakeven inflation rates. We now repeat all the steps of our analysis using higher frequency (daily) data, as well as lower frequency (quarterly) data.

The estimation results of oil price decomposition and breakeven inflation rates analysis (Equations (1) and (2), respectively) are summarized in Table 9. In Panel A we observe that pcAcmd has a positive and statistically significant coefficient in all three frequencies, and it explains 44-57 percent of the one-period percentage change in oil prices. In Panel B we observe that in all three frequencies the effect of the global aggregate demand factor embedded in oil prices (captured by the fitted value of Panel A, a1,pciAcmd) is higher in the post-crisis period. The effect of the remaining component (captured by the residual from panel A, Ut) is low throughout the sample period.

Image ppk

Image nvs

Image ino

Image tat

Image pus

Image jsp

Image fed

Image ius