Elsevier

Energy Economics

Volume 95, March 2021, 105127
Energy Economics

Asymmetric responses of consumer spending to energy prices: A threshold VAR approach

https://doi.org/10.1016/j.eneco.2021.105127Get rights and content

Highlights

  • We find asymmetries in the response of consumer spending to energy price shocks.

  • Positive shocks have a greater impact on consumer spending than negative shocks.

  • Our approach uses a threshold vector autoregression and U.S. data.

Abstract

We document asymmetric responses of consumer spending to energy price shocks: using a threshold vector autoregressive model estimated with Bayesian methods on U.S. data with high and low real energy inflation regimes, positive energy price shocks have a greater negative effect on consumption compared with the increase in consumption that arises from negative energy price shocks. For large shocks, the cumulative consumption responses are three to five times greater for positive than negative shocks. In disaggregated spending data, the asymmetric responses are strongest for durable goods consumption, but asymmetries are also present in the responses of nondurables and services consumption.

Introduction

The nominal price per barrel of crude oil fell by 65% from June 2014 to December 2015, and overall the prices of consumer energy goods and services as measured by the personal consumption expenditures price index fell by 23% over this period. A conventional view of the relationship between energy prices and consumer spending suggests that increases in oil prices slow consumer spending, akin to tax increases; in linear versions of this conventional view, a large decline in energy prices would be similar to a large tax cut, which would provide a strong boost to consumer spending. Thus, it is notable that—in real time—analysts, forecasters, and energy experts were generally surprised by the lack of a sizable response of consumer spending to the energy price declines; see, e.g., Hamilton (2015).1

There is a vast literature examining the relationship between the U.S. macro economy and oil price fluctuations (see, e.g., Hamilton 2005 and Kilian 2008 for surveys), and a key question in this literature has been whether the empirical evidence favors asymmetries and nonlinearities in this relationship or not. Surprisingly, while most of the impact of energy price shocks on economic growth is thought to flow through adjustments in consumer spending (e.g., Bernanke 2006), relatively little research has focused on potential asymmetries in the response of consumer spending to oil and energy prices. Using measures of positive oil price changes (following Mork 1989) or net oil price increases (following Hamilton, 1996, Hamilton, 2003), Mehra and Petersen (2005) present empirical evidence suggesting that consumer spending responds asymmetrically to oil price increases and decreases.2 More recently, Alsalman and Karaki (2019) use the nonlinear simultaneous equation model of Kilian and Vigfusson (2011a) to investigate the nonlinear relationship between consumer spending and oil prices, and find strong evidence of asymmetries and nonlinearities in the response of consumption to oil price shocks. By contrast, Edelstein and Kilian (2009) argue that one cannot reject the null hypothesis of symmetry in the impulse response functions of consumer spending to energy price shocks; thus, their analysis focuses on symmetric, linear models.

This paper provides empirical evidence of asymmetries in the response of consumer spending to energy price increases and decreases in the context of a multivariate, two-regime threshold vector autoregression (TVAR). TVARs provide a parsimonious and flexible methodology to investigate potential asymmetric and nonlinear relationships, as regimes endogenously evolve in response to changing conditions. Because the existence of multiple regimes raises the potential for overfitting due to parameter proliferation, we estimate the TVAR with Bayesian methods. Using monthly U.S. data on real consumption expenditures, energy inflation, and inflation excluding food and energy prices over an estimation sample spanning 1959–2014, the Bayesian measure of model fit—marginal likelihood—favors the two-regime model over a single regime, thus allowing for the possibility of nonlinearities and asymmetric responses of consumption to energy price shocks. The threshold variable separating our regimes is based on movements in real energy prices averaged over the past 9 months.3 Our posterior mean estimates put the threshold separating the two regimes at 0.36%. Such a threshold estimate is highly intuitive, as it allows for consumers to behave in two different fashions depending on whether real energy inflation is high (for practical purposes, positive) or low (for practical purposes, negative).

The estimated generalized impulse responses from our TVAR model show that consumer spending responds asymmetrically to positive and negative shocks to energy prices: qualitatively, the consumption responses are greater (in absolute terms) in response to positive energy price shocks than in response to the same-sized negative energy price shocks. Quantitatively, the asymmetries are greatest in response to large positive shocks and large negative shocks, in part because these shocks are most likely to push the economy into the high real energy inflation regime or the low real energy inflation regime, respectively, and to keep it there for some time within the context of our econometric model. At the posterior median, the consumption decline after one year in response to a 5% increase in energy prices is up to three times as large in absolute terms as the consumption response to a 5% decrease in energy prices, while the ratio increases to five in response to 10% increases and decreases in energy prices. Asymmetries are present but are more modest in the wake of small energy price shocks, in part because the economy remains near the threshold separating the regimes.4 Digging into the components of consumption, we find marked asymmetric responses of durable goods spending; asymmetric responses to nondurable goods spending and services spending are present but less pronounced.

The energy literature has noted a number of channels that could generate asymmetric economic responses to changes in energy prices, including an uncertainty channel that can cause consumers to postpone purchases of durables in response to heightened uncertainty from falling energy prices (Bernanke 1983) and a reallocation channel through which declines in energy prices have smaller impacts than increases in energy prices because the reallocation of resources in response to energy price declines offsets their economic benefits (Hamilton 1988). Energy price increases could also induce different responses than energy price decreases through broader behavioral explanations, for example by directly impacting consumer sentiment in different ways (Hamilton 2011), with relatively large energy price increases being particularly important for consumers and firms (Hamilton 1996). The asymmetries we document across components of consumer spending, and the greater evidence of asymmetries in response to large energy shocks, could also reflect an information channel. In particular, consumers may be better informed about large positive energy price shocks than about large negative energy price shocks, because there is usually greater news coverage of the former than of the latter (Knotek and Zaman 2021).

The paper is structured as follows. The next section describes the related literature. Section 3 describes the data and the empirical models. Section 4 discusses the empirical results. Section 5 reports the results for the components of consumption. Section 6 concludes.

Section snippets

Related literature

The vast majority of the literature that has investigated the relationship between energy prices (or oil prices) and aggregate macroeconomic variables has focused on industrial production, real GDP, real stock returns, unemployment, and payroll employment. With the exceptions of Mehra and Petersen (2005), Edelstein and Kilian (2009), Wang (2013), and Alsalman and Karaki (2019), which are the works most closely related to this paper, much less attention has focused on investigating the direct

Empirical models and data

Linear vector autoregressions (VARs) typically provide a useful starting point for assessing the economic impacts of energy price shocks; e.g., Edelstein and Kilian (2009) use a bivariate VAR in consumer spending and purchasing power changes related to energy price fluctuations. As a baseline, we use trivariate VARs where the vector Yt contains energy inflation, πtenergy, inflation excluding food and energy prices (core inflation), πtcore, and real consumption growth, Δ ln ct.11

Empirical results

To compare and contrast the linear VAR and TVAR models, we proceed by examining model fit and the plausibility of the estimated threshold, and then we move on to analyzing the impulse response functions and a historical decomposition exercise.

One advantage of estimating our models with Bayesian methods is the convenience of model comparisons in terms of model fit. Using the Bayesian metric of marginal likelihood (ML), we can rank model specifications in terms of the degree to which they fit the

Responses of consumption components and robustness

The estimated asymmetric and nonlinear behavior between aggregate consumption and energy prices may not necessarily hold for each of the components of aggregate consumption. To get a sense of which of the underlying components of the consumption basket may be contributing to the observed asymmetric behavior of aggregate consumption, we repeat the above exercises by examining the responses of the three main consumption components—consumer spending on durables, nondurables, and services—to energy

Conclusion

This paper reconsiders the issue of potential asymmetries and nonlinearities in the relationship between consumer spending and energy prices. Using a multivariate threshold vector autoregressive model estimated with Bayesian methods, we document that the historical U.S. data prefer multiple regimes over a single regime. Empirical estimates using our TVAR model show that consumption responses to positive and negative energy price shocks are consistently asymmetric: qualitatively, consumption

Declaration of Competing Interest

None

Acknowledgements

We thank Christiane Baumeister, Mark Bils, Oli Coibion, Jim Hamilton, Ana Maria Herrera, Kurt Lunsford, Rob Vigfusson, our discussants Nida Çakir Melek and Karel Mertens, two anonymous referees, and conference participants at the 5th IAAE Annual Conference, CEBRA Workshop for Commodities and Macroeconomics, the Federal Reserve Bank of Kansas City Day-Ahead System Energy Meeting, and the 25th Annual SNDE Conference for helpful comments, discussions, and suggestions. The views expressed herein

References (64)

  • Bwo-Nung Huang et al.

    The asymmetry of the impact of oil price shocks on economic activities: an application of the multivariate threshold model

    Energy Econ.

    (2005)
  • Gary Koop et al.

    A flexible approach to parametric inference in nonlinear and time varying time series models

    J. Econ.

    (2010)
  • Gary Koop et al.

    Impulse response analysis in nonlinear multivariate models

    J. Econ.

    (1996)
  • Mototsugu Shintani et al.

    Exchange rate pass-through and inflation: a nonlinear time series analysis

    J. Int. Money Financ.

    (2013)
  • Aditya Aladangady et al.

    Do Lower Gasoline Prices Boost Confidence?

  • Zeina Alsalman et al.

    Oil price shocks and the U.S. stock market: do sign and size matter?

    Energy Journal

    (2015)
  • Zeina Alsalman et al.

    Oil prices and personal consumption expenditures: does the source of the shock matter?

    Oxf. Bull. Econ. Stat.

    (2019)
  • Nathan S. Balke

    Credit and economic activity: credit regimes and nonlinear propagation of shocks

    Rev. Econ. Stat.

    (2000)
  • Nathan S. Balke et al.

    Oil price shocks and the U.S. economy: where does the asymmetry originate?

    Energy Journal

    (2002)
  • Marta Banbura et al.

    Large Bayesian vector auto regressions

    J. Appl. Econ.

    (2010)
  • Christiane Baumeister et al.

    Lower Oil Prices and the U.S. Economy: Is This Time Different?

    Brookings Papers on Economic Activity

    (Fall 2016)
  • Christiane Baumeister et al.

    Structural interpretation of vector autoregressions with incomplete identification: revisiting the role of oil supply and demand shocks

    Am. Econ. Rev.

    (2019)
  • Christiane Baumeister et al.

    Forty years of oil price fluctuations: why the price of oil may still surprise us

    J. Econ. Perspect.

    (2016)
  • Christiane Baumeister et al.

    Time-varying effects of oil supply shocks on the U.S. economy

    Am. Econ. J. Macroecon.

    (2013)
  • Ben S. Bernanke

    Irreversibility, uncertainty, and cyclical investment

    Q. J. Econ.

    (1983)
  • Ben S. Bernanke

    “Energy and the Economy” Speech before the Economic Club of Chicago, Chicago, IL

    (2006)
  • Andrea Carriero et al.

    Bayesian VARs: specification choices and forecast accuracy

    J. Appl. Econ.

    (2015)
  • Cathy W.S. Chen et al.

    Bayesian inference of threshold autoregressive models

    J. Time Ser. Anal.

    (1995)
  • Siddhartha Chib et al.

    Understanding the Metropolis-Hastings algorithm

    Am. Stat.

    (1995)
  • Ching-Wai Chiu et al.

    Macroeconomic tail events with non-linear Bayesian VARs

    (2016)
  • Todd E. Clark et al.

    Time variation in the inflation passthrough of energy prices

    Journal of Money Credit, and Banking

    (2010)
  • Luiggi Donayre et al.

    The asymmetric effects of oil price shocks on the Canadian economy

    Int. J. Energy Econ. Policy

    (2016)
  • Cited by (4)

    • The asymmetric effects of international oil prices, oil price uncertainty and income on urban residents’ consumption in China

      2022, Economic Analysis and Policy
      Citation Excerpt :

      Similarly, Wang (2013) asserted that rising oil prices have a more significant influence on household consumption than oil price decline; in addition, he found that the impact of oil price changes on household consumption has a threshold effect. Utilizing a threshold VAR model and U.S. data, Knotek II and Zaman (2021) concluded that positive energy shocks have a greater impact on consumer spending than negative shocks. However, Zhang et al. (2014) argued that, in China, while there is insufficient evidence to show that increasing oil prices negatively affect household consumption expenditures, it is clear that declining oil prices positively influence household consumption.

    View full text