The (time-varying) Importance of Oil Prices to U.S. Stock Returns: A Tale of Two Beauty-Contests.

AuthorBroadstock, David C.
  1. INTRODUCTION

    The probability distribution underlying a firm's expected cash flows can be altered by oil price fluctuations, either due to material impacts on future cash flows (creating effects that manifest through "production cost" and "income" channels), or on the discount rate as a result of uncertainty and subsequent interest rate changes ('uncertainty" and "discount rate" channels). This was the sentiment behind the pioneering study of Jones and Kaul (1996) who were among the first to argue that "stock returns... are negatively affected by both current and lagged oil price variables" (p.472).

    A vast body of literature has flourished around this conjecture, with the leading sentiment seeming to be that information about oil price changes can "help" to explain movements in financial markets (see, inter alia, Driesprong et al., 2008; Kilian and Park, 2009; or Narayan and Sharma, 2011). The evidence from this literature tends to point towards the following implications: (i) financial markets tend to respond negatively to increasing oil prices (Filis and Chatziantoniou, 2014; Asteriou and Bashmakova, 2013; Ciner, 2013); (1) (ii) stock markets in oil-exporting economies tend to respond positively to oil price increases (Arouri and Rault, 2012; Bjernland, 2009); (iii) there are asymmetric effects, where markets tend to respond differently to positive and negative changes in oil prices (Broadstock et al., 2016; Jimenez-Rodriguez, 2015; Broadstock et al., 2014); (2) (iv) there are asymmetric effects, where markets tend to respond differently to different oil price shocks depending whether these are demand-side or supply-side (Basher et al., 2018; Antonakakis et al., 2017; Kang et al., 2016); (3) (v) that effects are sector-specific and likely correlate with the degree of energy intensity of an industry (Degiannakis et al., 2013; Scholtens and Yurtsever, 2012; Arouri and Rault, 2012); (4) and (vi) that effects tend to be time-varying, with the reaction of stock returns to an unexpected shift in the price of oil not guaranteed to be the same as in preceding or following periods (Antonakakis et al., 2017; Boldanov et al., 2016). (5)

    There is a rich evidence base on the impact of oil prices, their changes, or some transformation/decomposition to obtain a measure of "oil price shocks," to stock market returns. Despite this, most prior research focuses either on aggregate stock market indices or selected industrial sectors. The existing literature has paid relatively little attention to examining the relationships using firm-level data, as noted for example by Antonakakis et al. (2018), Broadstock et al. (2016), Phan et al. (2015), Tsai (2015), Mohanty et al. (2013), Narayan and Sharma (2011), Sadorsky (2008), Boyer and Filion (2007) inter alia. Moreover, among the existing studies using firm-level data, only limited consideration has been given to the time-varying patterns of association between oil prices and stock markets. For example Broadstock et al. (2016) permitted only discrete time-variation captured through structural break tests, which allow a sudden discontinuity in the relationship, but which is far from being full time-varying.

    Against this backdrop we contribute to the literature by: (i) expanding the frameworks of Narayan and Sharma (2011) and Broadstock et al. (2016), insofar as we implement a "beauty contest" of sorts, sequentially analyzing a large sample of individual stocks; and more importantly (ii) placing special emphasis in testing for time-varying effects and relationships, placing our work in a unique position relative to the existing literature. For this purpose we employ a multi-factor empirical asset pricing framework in which the excess returns of a given stock i.e. the returns over and above the risk-free rate of return, are regressed against a number of price determining factors, including different measures of oil price fluctuations (i.e. to capture for example asymmetric or symmetric responses to oil price changes). Estimation is done using the dynamic model averaging (DMA) framework due to Koop and Korobilis (2012), which allows us to obtain time varying probabilities that oil price changes are among the set of indicators that govern excess stock return movements in any given period, as well as time-varying coefficients. In this paper, we place special emphasis on the interpretation of these probabilities.

    This topic is important for at least four reasons. First, given the recent shifts in international oil prices during 2014-2015, the sustained stagnancy at below-expected price levels through to the early part of 2016 and continuing into 2017, as well as, the upward movements since then, there is a valid question as to whether the role of oil prices to stock market performance has been altered. Second, it would be valuable to gauge how financial markets might react to the oil price recovery, which appears to be emerging towards the end of 2018 and generally continuing into the early part of 2019. Third, the majority of previous studies focus on aggregate or industrial indices and/or low frequency data, whereas in this study daily firm-level data are used, which provide richer information, (6) in the sense that heterogeneous effects are not masked by the very nature of the indices construction. Fourth, the majority of the studies concentrate on static econometric frameworks rather than time-varying frameworks. Hence, despite the wealth of literature, the question on how and whether oil prices impact stock market returns still remains open.

    Three unique aspects/contributions of this study include: (i) that we use a unique large-scale (bordering on "big") dataset comprising up to 25,372,588 observations measured at daily frequency from 10,118 U.S. listed firms, which to the best of our knowledge is the largest sample of firms/observations to be used for this purpose: (ii) that we place special emphasis on the time-varying nature of the oil price effects on excess stock returns, using an explicitly time-varying econometric framework; (iii) that we exploit the probabilistic information embodied within the model-averaging framework, allowing us to examine the time-varying "probability" that oil prices determine stock returns, albeit without describing the mechanisms that cause oil-related risk exposure (or resilience). These time-varying probabilities are central point of focus in our results section, and convey all the main insights we require. Nonetheless, we also summarize distributions for the key model coefficients.

    The central focus of our paper is around the application to daily data, however the popularity of oil price decompositions which isolate oil price changes due to supply or demand side effects e.g. the Kilian (2009) type decomposition, warrants additional analysis using monthly data. (7) To ensure the robustness and durability of our conclusions, we therefore re-implement all our main tests, with relatively minor modifications, to an equally extensive sample of data observed at monthly frequency comprising 6,320 individuals stocks with 1,448,294 firm-month observations. Noting that the reduced number of stocks stems from the need to keep a minimum number of observations to ensure we obtain reliable estimates.

    We phrase our analysis around the following hypotheses:

    Hypothesis 1: Oil price changes/oil price shocks have significant (important) impacts to the excess returns if individual firms.

    Hypothesis 2: The nature of the effect, and in particular the probability that excess returns are affected by oil price changes or shocks, is not constant over time i.e. it is time-varying. (8)

    Hypothesis 3(a): Excess returns exhibit heterogeneous responses to positive and negative oil price changes (asymmetry hypothesis).

    Hypothesis 3(b): Excess returns exhibit heterogeneous responses to oil price changes resulting from supply-side, or demand side price shocks (asymmetry hypothesis).

    Our findings offer convincing evidence that U.S. stocks generally do not contain a systematic premium due to oil price changes (symmetric, asymmetric) or shocks (demand or supply driven). We therefore lay challenge to the prevailing wisdom that oil-prices/shocks are important in determining the excess returns of listed firms in the US stock markets. We offer a generally more humbling appraisal of the topic, which supports the view that oil price fluctuations are not a "frequent" determinant of excess returns. This result is strongest when using daily frequency data, where, on average, stocks tend to price oil related information for around 2-3% of the sample period. For the monthly frequency data there is stronger evidence of the importance of oil price shocks, yet still the average stock is only significantly affected for 33% of the sample period, according to the results of our analyses. Reconciling our conclusions against the current consensus (that oil prices have wide reaching impacts over stock prices) we note that around 50-100% of stocks, depending on the use of daily or monthly frequency data, show a significant reaction to oil-price related information, at least once over the sample period. Regarding the hypothesized asymmetries, the evidence suggests for the daily data, the responses of excess stock returns to asymmetric oil price changes are more frequent compared to the symmetric changes. The picture is less clear for the monthly data, where excess stock returns seem to respond more frequently to different oil price shocks, depending on the time period. We should note, though, that this response is also at a fraction of the time over the sample period. Thus in summary the primary conclusion is that oil price shocks, though not negligible, are not a core determinant of excess stock returns.

    The remainder of the paper is structured as follows. Section 2 discusses the data used, Section 3 describes the econometric approach, while Section 4 reports the empirical findings...

To continue reading

Request your trial