Cross-Asset Time-Series Momentum Crude Oil Volatility and Global Stock Markets

Cross-Asset Time-Series Momentum:
Crude Oil Volatility and Global Stock Markets
Adrian Fernandez-Pereza, Ivan Indriawanb, Yiuman Tsec, Yahua Xud

a
Department of Finance, Auckland University of Technology, New Zealand
b
Adelaide Business School, University of Adelaide, Australia
c
Finance and Legal Studies Department, University of Missouri–St. Louis, United States
d
China Economics and Management Academy, Central University of Finance and Economics, China
August 2022
Abstract
We examine the profitability of a cross-asset time-series momentum strategy (XTSMOM)

constructed using past changes in crude oil implied volatility (OVX) and stock market returns
as joint predictors. We show that past changes in OVX negatively predict but past stock market
returns positively predict future stock market returns globally. The XTSMOM outperforms the
single-asset time-series momentum (TSMOM) and buy & hold strategies with higher mean
returns, lower standard deviations, and higher Sharpe ratios. The XTSMOM can also forecast
economic cycles. We contribute to the literature on cross-asset momentum spillovers as well
as on the impacts of crude oil uncertainty on stock markets.
Keywords: Cross-asset predictability; Crude oil market; International stock markets; OVX;
Time-series momentum
JEL classifications: G12; G15
Acknowledgments
We thank Geert Bekaert (the editor), two anonymous referees, Joelle Miffre and Michel Robe
for their helpful comments and suggestions. This study benefited from discussions with the
participants at the 37th International Conference of the French Finance Association (AFFI) and
the University of Technology Sydney finance seminar.
1
1. Introduction
In their seminal work, Jegadeesh and Titman (1993) show that, in US stock markets, past
winners (losers) continue to be winners (losers) over the next six months to a year. Similar
patterns have been documented in various asset classes, including international stocks
(Rouwenhorst, 1998), bonds (Jostova et al., 2013), commodities (Miffre and Rallis, 2007), and
currencies (Menkhoff et al., 2012). Moskowitz et al. (2012) introduce the time-series
momentum (TSMOM), in which the strategy is determined only by an asset’s past returns—
that is, prior one- to twelve-month returns positively predict future returns for different asset
classes (see also Goyal and Jegadeesh, 2018; Kim et al., 2016). In a recent study, Pitkäjärvi et
al. (PSV; 2020) propose a cross-asset time-series momentum (XTSMOM) strategy using past
bond and stock returns as predictors and find that the XTSMOM outperforms a single-asset
TSMOM strategy. Past bond returns are positive predictors of stock returns, whereas past stock
returns are negative predictors of bond returns.
In this paper, we incorporate the option-implied oil price volatility in XTSMOM to
predict stock markets around the world. Many empirical studies have shown the negative
effects of oil price uncertainty on economic growth and stock returns. Pindyck (1991) shows
that if an investment’s cash flow is dependent on the oil price, companies (particularly oil
companies) postpone irreversible investment decisions in response to increasing oil price
uncertainty. Elder and Serletis (2010) highlight that oil price uncertainty depresses current
investment. They report that oil price volatility has a negative effect on investment,
2
consumption, and aggregate output in the US. Jo (2014) finds similar results for global real
economic activity. Kellogg (2014) finds that oil and gas firms reduce their investment (proxied
using their oil drilling activity) when oil implied volatility increases. In addition, Kocaaslan
(2019) reports that oil price uncertainty significantly increases the US unemployment rate
while Nonejad (2020) shows that crude oil price volatility improves the short term forecast of
US real GDP growth rate.
The predictability of oil price uncertainty (measured by oil implied volatility) stock
markets can be explained by the funding constraints of financial intermediaries. Christoffersen
and Pan (2018) explain that being active in multiple markets, financial intermediaries finance
their trades using their own capital and/or collateralized borrowing from other intermediaries.
When oil-implied volatility is high, oil derivatives market margins increase and these
intermediaries become more capital constrained, leading to lower investment activity.
Therefore, an increase in oil implied volatility indicates a tightening in funding constraints,
which consequently affects the stock market. Consistent with this argument, Christoffersen and
Pan (2018, p.5) report that “after the financialization of commodity futures markets in 2004-
2005 oil volatility has become a strong predictor of returns and volatility of overall stock
market.” This reasoning is also consistent with Brunnermeier and Pedersen (2009) who show
that market uncertainty and funding liquidity are interrelated state variables of the stock market.
It is important to note that crude oil price uncertainty is influenced by many factors
including the overall uncertainty in the financial market. For instance, Cheng et al. (2015) use
3
changes in the Chicago Board Options Exchange Volatility Index (VIX) to proxy for shocks to
financial traders’ risk appetite and funding constraints. They show that financial traders reduce
their futures positions when they experience lower risk absorption capabilities due to a larger
exposure to the VIX. They further report a strong positive correlation between hedge fund
positions and oil (and other commodities) futures returns whereas commercial hedgers have a
negative correlation, consistent with the results of Büyükşahin and Robe (2014).1 If oil price is
influenced by the overall uncertainty in the financial market (proxied by the VIX), then oil
price uncertainty and the VIX should be closely related. The VIX could even be a key driver
of oil price uncertainty. Robe and Wallen (2016) find that the VIX captures much of what drives
uncertainty in the crude oil price, proxied by oil implied volatility. The power of the VIX
reflects the significant relation between equity and crude oil prices.2
Given all this evidence, we examine whether crude oil price uncertainty (measured by
the crude oil volatility index) can be employed as a trading signal for a profitable trading
strategy. We construct an XTSMOM strategy by examining the predictability of changes in the
CBOE Crude Oil Volatility Index (OVX) for various stock markets around the world. 3 Using
1
Cheng et al. (2015, p. 1733) call this convective risk flow: “Financial traders reduced their net long positions
during the crisis in response to market distress, whereas hedgers facilitated this by reducing their net short
positions as prices fell.” Büyükşahin and Robe (2014) find that hedge fund positions in commodity futures help
predict commodity-equity cross-market linkages. See also Büyükşahin and Robe (2010).
2
Robe and Wallen (2016) also show that macroeconomic variables contain little information regarding oil implied
volatility after controlling for the VIX, consistent with these results of Kilian and Vega (2011) who find that oil
price does not respond significantly to any US macroeconomic news within the day.
3
We employ the changes in OVX to identify positive vs. negative change in oil uncertainty because the OVX
4
OVX data over the sample period from May 2007 to August 2021, we first show that a cross-
asset momentum strategy using one-month lagged stock returns and OVX changes shows
superior predictability to a single-asset momentum strategy that relies on only one-month
lagged stock returns. More specifically, past stock returns can serve as positive predictors, and
past OVX changes can work as negative predictors for future stock returns. The results are
similar for oil-importing and oil-exporting countries. Therefore, investors cannot hedge the
effect of oil implied volatility shocks across the stock markets of oil importers and exporters.
We next evaluate the stock returns and Sharpe ratios under various TSMOM and
XTSMOM regimes. We find that the XTSMOM strategy with positive (negative) past stock
returns and negative (positive) past OVX changes generates the largest positive (negative)
future stock market returns and Sharpe ratios. We also compare the economic performance of
the XTSMOM with the TSMOM and buy & hold strategies. We document that the XTSMOM
has superior performance, with higher mean returns and higher Sharpe ratios than the two
alternatives. Furthermore, we find that the alpha of XTSMOM returns remains positive and
highly significant after controlling for Carhart’s (1997) four factors, the VIX, and its
components (Bekaert et al., 2013; Bekaert and Hoerova, 2014), the Amihud (2002) illiquidity
measure, the MSCI World index, the USD index, and value and momentum premiums across
several asset classes (Asness et al., 2013). These results are robust after accounting for
index is highly serially correlated. Furthermore, the OVX index is non-stationary whereas its first difference is
stationary. The use of first difference will remove the systematic pattern in the autocorrelation. This approach is
consistent with studies such as Ang et al. (2006), Bloom (2009) and Christoffersen and Pan (2018) who use also
the first difference in volatility as uncertainty shocks.
5
transaction costs and various adjustments to the stock return and crude oil volatility signals.
To provide supportive evidence of our economic story, we examine whether the lagged
changes in OVX can forecast funding liquidity proxies of financial intermediaries. We find that
funding constraints increase following an increase in oil uncertainty. This leads to lower
investment activities and stock market returns. Hence, the predictive power of the changes in
oil uncertainty on stock market returns can be explained by the funding liquidity argument of
Christoffersen and Pan (2018).
Finally, we explore the possible link between momentum strategies, such as the
TSMOM and the XTSMOM, and the real economy by focusing on key economic indicators
such as the industrial production index, the unemployment rate, and the inflation rate. We find
that the XTSMOM can predict economic cycles. In particular, the long (short) XTSMOM
portfolio—namely, positive (negative) stock returns and negative (positive) OVX changes—
can point to good (bad) economic times, with higher (lower) industrial production, decreasing
(increasing) unemployment rates, and lower (higher) inflation.
It is important to note that the response of stock returns to oil price shocks can be
positive or negative, depending on the source of the oil price shock. We follow Kilian (2009)
to decompose oil price shocks into supply, aggregate demand, and oil-specific demand shocks.
We observe that supply and oil-specific demand shocks decrease XTSMOM returns, but
aggregate demand shock does not. Our findings indicate that the XTSMOM performance is
only partially explained by the various oil shocks.
6
Our study extends the literature on the global impacts of oil uncertainty (e.g., Guo and
Kliesen, 2005; Kwon, 2020; Gao et al., 2022). We find that OVX changes can serve as a
significant predictor of stock markets. We complement the results of Christoffersen and Pan
(2018), who highlight the importance of funding constraints, by incorporating XTSMOM into
the predictability of oil implied volatility for global stock returns.
We organize the remainder of the paper as follows. We discuss the data in Section 2.
Section 3 explains the methodology and reports the empirical results. We conduct further
analyses in Section 4. Section 5 concludes.
2. Data and Descriptive Statistics
2.1. Data Sources
We download crude oil, stock market, and economic data at a monthly frequency from various
sources. As a measure of crude oil volatility, we use the Crude Oil Exchange Traded Fund (ETF)
Volatility Index (OVX) obtained from Refinitiv Datastream. The OVX is an estimate of the
expected 30-day volatility of crude oil as priced by the United States Oil Fund (USO). The
Chicago Board of Exchange (CBOE) first published the OVX on May 9, 2007. Thus, we start
our sample period in May 2007 and end in August 2021.
For the stock market data, we obtain the monthly MSCI total return index for various
countries from Refinitiv Datastream.4 To remove the effect of exchange rate differences, we
4
The Morgan Stanley Capital International (MSCI) stock index is often tracked by exchange traded funds (ETFs)
7
download all equity indices in USD. Third, we collect the US short-term (one-month) interest
rates from Kenneth French’s website. For this study, we consider a US investor who takes a
position in international stock markets to exploit the XTSMOM strategy and invests in the US
risk-free rate when not taking positions. In total, we consider 59 countries, including 44 oil-
importing and 15 oil-exporting countries. Our final sample consists of 8,220 country-month
observations.5
For each country, we download from Refinitiv Datastream the MSCI small- and large-
cap indices to construct the SMB (small minus big) risk factor (calculated as the MSCI small-
minus the MSCI large-cap returns). We also download the MSCI value and growth indices to
construct the HML (high minus low) risk factor (calculated as the MSCI value minus the MSCI
growth returns). To proxy for stock market risk, we use the VIX obtained from Refinitiv
Datastream. Bekaert, Hoerova, and Lo Duca (2013) explain that the VIX has two components:
conditional variance (which captures uncertainty) and the variance premium (which captures
risk aversion). We obtain data on the VIX components from Marie Hoerova’s website.
In our study, we also examine whether the impact of oil price shocks on stock markets
differs depending on whether the change in the price of oil is driven by demand or supply
shocks. As such, we extract the crude oil shocks from the following datasets: (1) Kilian’s Global
and futures contracts in different countries (Angelidis and Tessaromatis, 2017). Hence, using the MSCI stock
market index is realistic to ensure that the XTSMOM strategy is implementable.
5
The list of these countries and their stock index symbols is in Appendix A.
8
Economic Activity as an indicator of the aggregate oil demand,6 (2) the change in world crude
oil production, including lease condensate, as an indicator of the change in global crude oil
supply,7 (3) the US crude oil composite acquisition cost by refiners8 deflated by US CPI9 as an
indicator of the real crude oil price. We collect monthly data and extract the crude oil shocks
from January 1974 to August 2021 and match them with our sample period for our regression
analysis.
Finally, we collect information on several economic fundamentals. Like PSV (2020),
we focus on industrial production, the unemployment rate, and the inflation rate, as they are
consistently available across the countries in our sample. These macroeconomic data are
collected at a monthly frequency from Refinitiv Datastream.
2.2. Descriptive Statistics
Table 1 reports the statistics of the OVX changes and the pooled stock excess returns for all the
countries and for oil-importing and oil-exporting countries. Panel A reports that the changes in
OVX is only slightly positive over the sample period, with an average of 0.05% per month (t-
statistic of 0.07) and a standard deviation of 12.18% per month.10 The OVX changes are also
6
https://www.dallasfed.org/research/igrea/.
7
https://www.eia.gov/international/data/world/.
8
https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET&s=R0000_3&f=M/.
9
https://www.bls.gov/cpi/.
10
Our sample has more months with negative than positive OVX changes. Specifically, 54.4% of our sample
9
positively skewed and have large tails (i.e., high kurtosis). With respect to the pooled stock
returns, we observe positive monthly mean excess returns at around 0.417% per month for all
countries, 0.392% per month for oil-importing countries, and 0.494% per month for oil-
exporting countries. On average, the pooled stock returns are negatively skewed and have large
tails. The last column suggests that the OVX changes and equity excess returns are not
persistent, with first autocorrelation (AR(1)) coefficients ranging from -0.057 (for the OVX
changes) and 0.093 (for oil-exporting countries excess returns).
[Insert Table 1 here]
3. Empirical Results
In this section, we first investigate whether the signs of past OVX changes can predict future
stock returns in an international context. We then show that these signals can be exploited in
an XTSMOM strategy. Next, we provide the economic rationale for the predictive power of
past OVX changes predictability and study how the XTSMOM returns relate to oil
supply/demand shocks. Finally, we report the robustness of our results.
3.1. Single- and Cross-Asset Time-Series Predictability
Before we apply the cross-asset time-series predictability model, we start by explaining the
period has decreases in the OVX, 44.4% increases in the OVX, and another 1.2% has no changes. In an unreported
result, we observe some clusters of positive OVX change signals that coincide with financial crises, such as the
global financial crisis (2008), the 2014 Russian financial crisis (second half of 2014), and the COVID-19
pandemic (early 2020).
10
single-asset time-series predictability model. In particular, we can assess whether the signs of
past stock market returns are predictive of future returns using the following pooled panel
regression:
𝑟𝑡𝑒,𝑖 = 𝛼 + 𝛽 𝑒 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) + 𝜀𝑡𝑖 , (1)
where 𝑟𝑡𝑒,𝑖 is the stock market excess returns of country i in month t, 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) is the sign of
the previous month market i excess returns, {𝛼, 𝛽 𝑒 } are the parameters to estimate, and 𝜀𝑡𝑖 are
the residuals. This single-asset model has been used in various studies, such as Moskowitz et
al. (2012), Kim et al. (2016), and Goyal and Jegadeesh (2018). In our study, we focus on a
lookback period of one month. However, we also use the average past stock excess return over
the previous year instead of a one-month lag for robustness. We now extend our analysis of
time-series predictability by examining whether the signs of lagged OVX changes (∆𝑂𝑉𝑋𝑡−1 )
are predictive of future stock market returns. To do so, we pool all stock market returns and
regress them on the sign of lagged OVX changes as well as the sign of its own lagged excess
returns as follows:
𝑟𝑡𝑒,𝑖 = 𝛼 + 𝛽 𝑒 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) + 𝛽 𝑂𝑉𝑋 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−1 ) + 𝜀𝑡𝑖 . (2)
The coefficients {𝛼, 𝛽 𝑒 , 𝛽 𝑂𝑉𝑋 } are estimated using a pooled panel regression.11 In Table 2, we
11
Before estimating Equation (2), we examine the associations between the equity and crude oil markets. More
specifically, we applied a panel Granger-causality test using the Dumitrescu-Hurlin approach on the equity return
and OVX change series. This approach runs standard Granger-causality regressions for each country individually.
11
report the results of Equation (2) along with the White-corrected t-statistics clustered by year-
month. For comparison, we also report the results of Equation (1) with the sign of the one-
month lagged stock market returns as a predictor. We present the results for all countries (Panel
A), oil-importing countries (Panel B), and oil-exporting countries (Panel C).
Panel A shows that the predictability of the sign of the lagged OVX changes is negative
and significant at the 1% level for all countries (t-statistic of –3.59). This result suggests that
stock returns are 1.00% per month lower when the one-month lagged OVX change is positive,
i.e., when the crude oil volatility increased in the previous month. Adding the one-month lagged
OVX change increases the adjusted R2 by 1.46% compared to the single-asset model with its
own one-month lagged stock returns.12 To ensure that these results hold across different periods
in our sample, we re-estimated Equation (2) using two equal-sized subsamples, June 2007 –
July 2014 and August 2014 – August 2021. In both cases, the negative predictability of
𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋) persists. These results are available from the authors.
It then takes the average of the test statistics, standardized to follow a standard normal distribution. We find that,
regardless of the number of lags employed (from one to five), OVX changes Granger cause equity returns, but
not the reverse. It therefore supports our motivation for the cross-asset time-series momentum effect between
crude oil volatility and stock markets.
12
In unreported results, we employed alternative lags for the stock excess return and OVX changes in Equations
𝑒,𝑖
(1) and (2), i.e., 𝑠𝑖𝑔𝑛(𝑟𝑡−ℎ ) and 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−ℎ ) with h=1, …, 12, respectively. The results show positive
predictability from past stock excess returns up to the fifth lag, with a reversal afterward; however, none of the
lags are statistically significant. Otherwise, the sign of the first lag of OVX change is the only statistically
significant lag. This result can be explained by how the OVX index is obtained. Specifically, the OVX represents
the market expectations about future oil volatility at the one-month horizon, therefore, its predictability is stronger
at a one-month horizon. These results are available upon request from the authors.
12
Robe and Wallen (2016) document that the VIX is one of the main drivers of the market
expectations of crude oil implied volatility. Therefore, the OVX is closely linked to the VIX.
To examine whether the OVX changes drive stock market returns beyond the VIX or its
components, we include the sign in VIX changes (𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋)) and the sign in the changes in
its two components (𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝐶𝑉)and 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃)). In addition, we control for stock
market predictors, such as size (SMB) and book-to-market (HML) risk factors. We observe that
part of the predictive ability of XTSMOM is explained by the control variables. However, the
sign of the one-month lagged OVX change remains negative and statistically significant,
suggesting that the crude oil volatility index provides a trading signal beyond that of stock
market information.
Panels B and C of Table 2 further show that the lagged OVX changes predictability is
negative and significant for oil-importing and oil-exporting countries, with t-statistics of
similar magnitude for the coefficients and adjusted R2. In all three panels, we observe that
𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋) and the signs of the two components of VIX are insignificant at the 10% level,
but 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃) is marginally significant at the 10% level in Panels A and C.
Although positive oil price changes have negative impacts on oil-importing economies
and positive impacts on oil-exporting countries (Barsky and Kilian, 2004; Wang et al., 2013),
the effect of changes in oil implied volatility is of similar magnitude and sign in both groups.
This finding is consistent with Diaz et al. (2016) who find a negative response by the G7 stock
13
markets (including Canada, an oil-exporting country) to an increase in world oil price volatility.
Our finding is also consistent with Wang (2021) who document the propagation of oil price
shocks through banks’ operations including a decline in demand deposit, a surge in credit line
drawdowns and a jump in troubled loans, particularly for banks with significant operations in
oil-concentrated countries. We therefore conclude that oil implied volatility generally has a
negative impact on all our sample countries.
Appendixes B and C report the results of Equation (2) by country. The negative
relationship between the sign of the one-month lagged OVX changes and the country’s stock
return is persistent across countries. Among our 59 countries, 58 have a negative 𝛽 𝑂𝑉𝑋
coefficient, and they are statistically significant at the 10% level or better in 30 countries (about
51% of the countries in our sample).
In our next analysis, we assess the economic value of the XTSMOM strategy. We report
the average monthly excess returns and Sharpe ratios by the stock market and OVX change
regimes in Table 3. We present the results for the stock returns and OVX changes momentum
regimes separately in Panel A. In Panel B, we use both stock returns and OVX changes as joint
predictors.
Panel A shows that stock returns are highest (lowest) during positive (negative) stock
momentum regimes. Stock returns are also the highest (lowest) during negative (positive) OVX
changes. These findings confirm the regression results in Table 2 and are consistent with past
14
stock returns being positive predictors of future stock returns and past OVX changes being
negative predictors of future stock returns. In Panel B, we show that a profitable trading
strategy is formed from combining the lagged stock returns and OVX changes as joint
predictors. The highest stock excess returns and Sharpe ratios are observed in months with
positive past stock excess returns combined with negative past OVX changes. In contrast, the
lowest stock excess returns and the lowest Sharpe ratios are observed in months with negative
past stock returns combined with positive past OVX changes.
3.2. Cross-Asset Time-Series Strategy
Previously, we showed that past OVX changes are a significant predictor of future stock market
returns in an international context, and this predictability is not subsumed by the information
of the lagged stock market return and various control variables. In this section, we exploit this
pattern in a cross-asset trading strategy. First, we explain how to construct the single-asset
TSMOM strategy. Then we describe how we modify it to construct the XTSMOM strategy.
The single-asset TSMOM strategy takes a long (short) position in the stock market
index of country i in month t if its past-month excess return is positive (negative). The position
is held for one month, after which the previous steps are repeated. Specifically, the single-asset
TSMOM strategy excess return, 𝑟𝑡𝑇𝑆𝑀𝑂𝑀,𝑖 , is calculated as follows:
𝑟𝑡𝑇𝑆𝑀𝑂𝑀,𝑖 = 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) ∙ 𝑟𝑡𝑒,𝑖 . (3)
15
We form a diversified TSMOM portfolio with this single-asset TSMOM return series by taking
an equal-weighted average of the individual assets’ TSMOM returns—that is, 𝑇𝑆𝑀𝑂𝑀𝑡 =

𝑇𝑆𝑀𝑂𝑀,𝑖
∑𝑁
𝑖=1 𝑟𝑡
, where i denotes the country, and N is the total number of countries in the sample.13
𝑁
The XTSMOM strategies are constructed by incorporating the cross-asset predictor as
well as the single-asset time-series predictor. More specifically, we take a long (short) position
in a given month if the past monthly stock market excess return is positive (negative) and that
of the OVX change is negative (positive). Otherwise, we hold a risk-free asset—that is, the
one-month US interest rate. We hold the position for one month and then repeat the previous
steps. The XTSMOM excess returns, 𝑟𝑡𝑋𝑇𝑆𝑀𝑂𝑀,𝑖 , for country i in month t are obtained as
follows:
+𝑟𝑡𝑒,𝑖 , 𝑖𝑓 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) > 0 and 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−1 ) < 0
𝑟𝑡𝑋𝑇𝑆𝑀𝑂𝑀,𝑖 = {−𝑟 𝑒,𝑖 , 𝑖𝑓 𝑠𝑖𝑔𝑛(𝑟 𝑒,𝑖 ) < 0 and 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−1 ) > 0 (4)
𝑡 𝑡−1
0, otherwise.
We form diversified cross-sectional TSMOM portfolios, denoted XTSMOM, by taking equal-
weighted averages of the individual assets’ XTSMOM returns—that is, 𝑋𝑇𝑆𝑀𝑂𝑀𝑡 =

𝑋𝑇𝑆𝑀𝑂𝑀,𝑖
∑𝑁
𝑖=1 𝑟𝑡
, where i denotes the country, and N is the total number of countries in the sample.
𝑁
13
Disentangling the TSMOM from risk parity investment is not trivial (see, e.g., Kim et al., 2016). We therefore
follow PSV (2020) and do not apply volatility scaling in the TSMOM strategy.
16
Table 4 reports the performance of the XTSMOM and TSMOM strategies together with
the buy & hold strategy, which we use as a benchmark.14 We report the performance for all the
countries (Panel A), for oil-importing countries (Panel B), and oil-exporting countries (Panel
C). In Panel A, the buy & hold strategy obtains a positive mean excess return of 0.41% per
month (5.03% per year) but is statistically insignificant. The TSMOM strategy obtains a
positive mean excess return of 0.52% per month (6.42% per year), also statistically
insignificant.
Notably, the XTSMOM outperforms all the strategies, with a mean excess return of
0.79% per month (9.90% per year), significant at the 5% level, and a monthly Sharpe ratio of
0.175. We test for the difference in mean returns using a t-test of differences in means between
XTSMOM versus buy & hold (𝐻1 : 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀 − 𝑀𝑒𝑎𝑛𝐵&𝐻 = 0 ) and XTSMOM versus
TSMOM (𝐻2 : 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀 − 𝑀𝑒𝑎𝑛𝑇𝑆𝑀𝑂𝑀 = 0 ). We reject the null hypothesis in both cases
at the 5% level with t-statistics of 2.30 (𝐻1 ) and 2.07 (𝐻2 ), confirming that the XTSMOM mean
return is economically and statistically superior to those of the two alternatives. We also test
the significance of the Sharpe ratios using Opdyke’s (2007) Sharpe ratio test with a null
hypothesis that the Sharpe ratio is equal to zero (𝐻3 : 𝑆𝑅𝑗 = 0 for any strategy j). We reject the
null hypothesis at the 1% level (p-value of 0.004) for the XTSMOM strategy, suggesting that
14
The buy & hold strategy invests equally in all the countries and maintains a long position in our holding period,
which is one month. The same process is repeated for the subsequent periods.
17
not only the mean return but also the risk-adjusted performance of XTSMOM is statistically
and economically significant. The XTSMOM portfolio also has lower risk, with a lower
standard deviation and similar kurtosis but more positive skewness than those of the TSMOM.
These results are consistent for both the oil-importing (Panel B) and oil-exporting (Panel C)
countries. We also test the difference in mean returns for the XTSMOM strategy between the
oil-importing and oil-exporting countries in Panels B and C, respectively
( 𝐻4 : 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀(𝑒𝑥𝑝𝑜𝑟𝑡𝑖𝑛𝑔) − 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀(𝑖𝑚𝑝𝑜𝑟𝑡𝑖𝑛𝑔) = 0 ). Consistent with our earlier
results, we do not find any statistical difference between the mean returns, with a t-statistic of
1.44.15
Figure 1 plots the future value of $1 invested with the buy & hold, TSMOM, and
XTSMOM strategies. Consistent with Table 4, the XTSMOM strategy outperforms both
TSMOM and buy & hold, with final earnings of $3.61 at the end of our sample period. In
comparison, the TSMOM and buy & hold strategies have final earnings of $2.26 and $1.62,
respectively. We also consider alternative investment dates, i.e., $1 invested from May 2007
until June 2014 and $1 invested from June 2014 until August 2021. The plots in Appendix D
show that the XTSMOM strategy yields a higher profit than the buy & hold and TSMOM
strategies in both subperiods. These findings suggest that the cross-asset time-series
15
It is probable that the US stock market better predict other stock markets, and therefore, yields a more
challenging benchmark (Rapach et al., 2013). We tested this by replacing the lagged local stock returns with the
lagged US stock returns as the equity signal in the XTSMOM strategy. We find that using local stock market
return as signals offers a better performance (Sharpe ratio of 0.175) compared to using US stock returns (Sharpe
ratio of 0.151). The results are available from the authors upon request.
18
momentum strategy remains profitable, regardless of the initial investment date.
[Insert Figure 1 here]
Figure 2 plots the Sharpe ratios of the XTSMOM and TSMOM strategies for each
country in our sample. In general, the Sharpe ratios of the former are superior to those of the
latter in all countries except Bulgaria and Chile. Hence, we conclude that the XTSMOM
strategy is superior not only in the context of an international stock market portfolio but also
in most countries.
We further analyze the excess returns of the XTSMOM strategies by calculating their
alphas based on the following models:
𝑋𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + 𝛽1 𝑇𝑆𝑀𝑂𝑀𝑡 + 𝛽2 𝑀𝐾𝑇𝑡 + 𝛽3 𝑆𝑀𝐵𝑡 + 𝛽4 𝐻𝑀𝐿𝑡 + 𝛽5 𝑈𝑀𝐷𝑡 + 𝛽6 ∆𝑉𝐼𝑋𝑡 +
𝛽7 ∆𝐼𝐿𝐿𝐼𝑄𝑡 + 𝜖𝑡 , (5)
𝑋𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + 𝛽1 𝑇𝑆𝑀𝑂𝑀𝑡 + 𝛽2 𝑀𝑆𝐶𝐼_𝑊𝑜𝑟𝑙𝑑𝑡 + 𝛽3 𝑉𝐴𝐿_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 +
𝛽4 𝑀𝑂𝑀_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 + 𝛽5 ∆𝑢𝑠𝑑𝑡 + 𝜖𝑡 , (6)
where 𝑋𝑇𝑆𝑀𝑂𝑀𝑡 denotes the excess return in month t of the XTSMOM portfolio, and
𝑇𝑆𝑀𝑂𝑀𝑡 denotes the excess return for the TSMOM portfolio. In Equation (5), we include the
three factors of Fama and French (1993), {𝑀𝐾𝑇𝑡 , 𝑆𝑀𝐵𝑡 , 𝐻𝑀𝐿𝑡 }, and the momentum factor of
Carhart (1997), 𝑈𝑀𝐷𝑡 , for developed markets collected from Kenneth French’s website. We
19
also include ∆𝑉𝐼𝑋𝑡 , which is either the VIX changes or the changes in its components, and the
changes in the Amihud (2002) illiquidity ratio, ∆𝐼𝐿𝐿𝐼𝑄𝑡 ,16 as a measure of liquidity risk. In
Equation (6), 𝑀𝑆𝐶𝐼_𝑊𝑜𝑟𝑙𝑑𝑡 is the MSCI World returns (in USD) collected from Refinitiv
Datastream.17 𝑉𝐴𝐿_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 and 𝑀𝑂𝑀_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 are the cross-sectional value and
momentum returns obtained across asset classes from the AQR data library used by Asness et
al. (2013). We also control for foreign exchange risk through the log changes in the USD index
downloaded from Refinitiv Datastream.18 Finally, 𝜖𝑡 denotes the residuals.
Table 5 reports the regression estimates from Equation (5). In Panel A, we show that
using all the countries, the XTSMOM strategy obtains a sizable abnormal performance or alpha
of 0.413% per month (4.96% per year), on average, and is statistically significant at the 5%
level. The control variable TSMOM is significant at the 1% level, while other market risk
factors, such as HML, UMD, ∆𝑉𝐼𝑋 and ∆𝑉𝐼𝑋_𝑉𝑃, are statistically significant at the 10% level
or better. These coefficients suggest that the countries in the long (short) XTSMOM portfolio
tend to be growth (value), losers (winners), with high (low) uncertainty and risk aversion.
Similar results are observed for the oil-importing (Panel B) and oil-exporting (Panel C)
16
The Amihud illiquidity ratio captures the price impact of trading. The illiquidity measure for month t is
1 |𝑟𝑑𝑈𝑆 |
calculated as 𝐼𝐿𝐿𝐼𝑄𝑡 = ∑𝐷
𝑑=1 𝑈𝑆 , where 𝑟𝑑𝑈𝑆 is the stock return on day d for the daily MSCI US index,
𝐷 $𝑉𝑜𝑙𝑑
$𝑉𝑜𝑙𝑑𝑈𝑆 is the daily dollar trading volume in the US, and D is the total number of trading days in a month. Data
are collected from Refinitiv Datastream.
17
MSCI World Index comprises large- and mid-cap stock performance across 23 developed countries.
18
The US dollar (USD) index is a measure of the value of the US dollar relative to the value of a basket of six
world currencies—the euro, Swiss franc, Japanese yen, Canadian dollar, British pound, and Swedish krona.
20
countries.
The results for Equation (6) are reported in Table 6. In Panel A, we observe that the
alpha from the XTSMOM strategy is positive at 0.46% per month (5.69% per year) and
significant at the 1% level. This outperformance persists even after world stock returns, cross-
sectional value and momentum returns, and the log changes in the USD index are taken into
account.
Finally, we implement a spanning test in which we regress the monthly returns of the
XTSMOM and TSMOM portfolios on each other. The results in Appendix E confirm that the
XTSMOM is not equivalent to the TSMOM. For instance, Panel A shows that the XTSMOM
alpha is still positive (0.36% per month, on average) and significant at the 5% level when we
control for the TSMOM. It is important to note that the TSMOM is negative but insignificant
when we control for the XTSMOM. In other words, the information captured by the TSMOM
is included in the XTSMOM, but not the reverse.
3.3. Funding constraints of financial intermediaries
In this study, we argue that changes in oil price volatility may predict stock market returns due
to the funding constraints of financial intermediaries. To provide supportive evidence of our
economic story, we have regressed several funding liquidity proxies on one month lagged
changes in OVX, similar to Christoffersen and Pan (2018). If changes in OVX affect funding
21
liquidity of financial intermediaries, it should be able to forecast the funding liquidity proxies.
More specifically, we estimate the following regression:
𝐹𝐿𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦𝑡 = 𝛼 + 𝛽1 ∆𝑂𝑉𝑋𝑡−1 + 𝜖𝑡 , (7)
where 𝐹𝐿𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦𝑡 is the measure of funding constraints of financial intermediaries in month
𝑡. We consider four funding liquidity proxies: (1) the TED spread from St. Louis Fed (obtained
as the difference between the 3-month LIBOR and 3-Month Treasury bills), (2) the credit
spread from St. Louis Fed (the difference between Baa and 10-year constant maturity Treasury
bonds), (3) the Betting-against-Beta (BAB) factor of Frazzini and Pedersen (2014) from AQR,
and (4) the International Bank index returns from Refinitiv Datastream. Increases in both the
TED and credit spreads reflect funding illiquidity. If an increase in oil implied volatility
decreases funding liquidity, we expect the regression coefficients for the lagged OVX changes
to be positive. The BAB factor measures returns of a portfolio that is long low-beta stocks and
short high-beta stocks. Low BAB factor signals funding constraints. Hence, if an increase in
oil implied volatility decreases funding liquidity, we expect the regression coefficients to be
negative. Finally, International Bank index is an equally weighted stock price index of large
global commercial banks. Decreases in the bank index return imply that broker–dealers are
more constrained and/or face higher margin requirements. Thus, if an increase in oil implied
volatility decreases funding liquidity, we expect the regression coefficients to be negative.
Table 7 reports the estimates, Newey-West corrected t-statistics and adjusted-R2. In line
with Christoffersen and Pan (2018), we find that an increase in OVX changes increases future
TED and credit spreads. Although the coefficients have the expected sign, they are not
statistically significant. We also find that an increase in OVX changes decreases the BAB factor
and International Bank index returns in the next period. Their coefficients have the expected
22
sign and are significant at the 1% and 10% level, respectively. These results provide supportive
evidence that the economic linkage between the changes in oil implied volatility and future
stock markets returns can be explained by the financial intermediary channel, consistent with
Christoffersen and Pan (2018).
INSERT TABLE 7 HERE
3.4. Oil Supply and Demand shocks
Kilian and Park (2009) show that the impact of oil price shocks on US stocks differs, depending
on whether the change in the price of oil is driven by demand or supply shocks in the oil market.
To examine whether the performance of XTSMOM is explained by such shocks, we first
extract the oil shocks using the structural VAR (SVAR) model developed by Kilian (2009).
More specifically, we identify oil supply, global demand, and oil-specific demand shocks using
world crude oil production (as a proxy for oil supply), Kilian’s global economic activity (as a
proxy for global oil demand), and the deflated US crude oil composite acquisition cost by
refiners (as a proxy for the real price of oil).19 We then estimate the following regression to
assess the effect of shocks to crude oil on the performance of the XTSMOM strategy,
𝑋𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + 𝛽1 𝑆𝑢𝑝𝑝𝑙𝑦𝑡 + 𝛽2 𝐴𝑔𝑔_𝑑𝑒𝑚𝑎𝑛𝑑𝑡 + 𝛽3 𝑂𝑖𝑙_𝑑𝑒𝑚𝑎𝑛𝑑𝑡 + 𝜖𝑡 , (8)
19
The structural VAR results are available upon request from the authors.
23
where 𝑋𝑇𝑆𝑀𝑂𝑀𝑡 are the excess returns of the XTSMOM strategy at month t, 𝑆𝑢𝑝𝑝𝑙𝑦 is the
oil supply shock, 𝐴𝑔𝑔_𝑑𝑒𝑚𝑎𝑛𝑑𝑡 is the aggregate demand shock, 𝑂𝑖𝑙_𝑑𝑒𝑚𝑎𝑛𝑑𝑡 is the oil-
specific demand shock, and 𝜖𝑡 is the residual.
The results reported in Panel A of Table 8 show that supply and oil-specific demand
shocks have negative and significant coefficients. This finding suggests that an unexpected
increase in oil production (which causes a decrease in the real price of oil) or an unexpected
increase in the precautionary demand for crude oil (which causes an increase in the real price
of oil) decreases the XTSMOM returns. We do not find the aggregate demand shock coefficient
to be statistically significant, suggesting that an unexpected change in aggregate demand does
not explain the performance of the XTSMOM strategy. More importantly, the constant remains
positive and statistically significant. These results are consistent across the all-country sample,
as well as for oil-importing (Panel B) and oil-exporting (Panel C) countries.
The Theory of Storage (Working, 1933; Gorton et al., 2012) links levels of inventories
with price volatility. High inventories are associated with a decrease in price volatility, and low
inventories with an increase in price volatility. In line with this argument, we interpret the
negative coefficient for oil-specific (or precautionary) demand in Table 5 as the return of
XTSMOM strategy being lower during periods of low oil uncertainty, i.e., when precautionary
demand for oil is high. Similarly, an increase in oil supply reduces oil uncertainty. As such, oil
supply shocks lead to lower profitability for the XTSMOM strategy. To test the above claim,
we split our sample into periods of high oil uncertainty (i.e., when the OVX is higher than its
full sample mean) and periods of low oil uncertainty (i.e., when the OVX is lower than its full
sample mean). The Sharpe ratio of the XTSMOM strategy for all countries in periods of high
oil uncertainty (0.283) is superior to that in periods of low oil uncertainty (0.040). The results
24
suggest the XTSMOM strategy performs strongly during periods of high oil uncertainty.20
3.5. Robustness to the XTSMOM Strategy
We perform several robustness tests of the XTSMOM strategy based on the following: (1)
using the past 12-month mean stock return as the stock return signal, (2) using the difference
in the level of OVX and its 12-month moving average as the oil uncertainty signal, (3) after
accounting for transaction costs and short-selling constraints, (4) using the stock market excess
returns in local currency, (4) using an alternative scale of the XTSMOM returns, and (5) after
orthogonalizing the OVX series.
3.5.1. Average 12-Month Past Returns as the Stock Market Signal
So far, we observe that the TSMOM offers inferior performance to the XTSMOM. This might
be because the past-month stock return is a poor and noisy signal for future stock returns. To
circumvent this issue, the literature suggests employing the average of a larger window as the
stock return signal in a TSMOM strategy (see, e.g., Moskowitz et al., 2012; PSV, 2020). In this
20
In a further analysis, we study which type of oil shocks give rise to the oil uncertainty predictability. We regress
the OVX changes on the contemporaneous oil supply and demand shocks. We find that OVX changes are
contemporaneously affected by oil-specific (or precautionary) demand shocks. The importance of the oil-specific
demand shocks is in line with Kilian (2009) who states, “oil price shocks historically have been driven mainly by
a combination of global aggregate demand shocks and precautionary demand shocks, rather than oil supply shocks,
as is commonly believed.” In other words, a positive oil-specific demand shock is associated with an increase in
the inventory of oil and a decrease in oil implied volatility. Therefore, we conclude that the predictability from
OVX changes is partially explained by the oil-specific demand shocks.
25
robustness test, we use the average of the previous 12-month excess returns as the stock market
signal in the XTSMOM strategy, keeping the one-month lagged OVX change as the signal of
oil implied volatility.
Figure 3 shows the robustness results. The second set of bar charts shows that using the
average of the previous 12-month return as the stock market signal yields a monthly Sharpe
ratio of 0.145 (XTSMOM), 0.004 (TSMOM), and 0.069 (buy & hold). Although the magnitude
is smaller than the baseline result, XTSMOM still outperforms the TSMOM and buy & hold
strategies.
3.5.2. Alternative Oil Uncertainty Signal
As an alternative oil uncertainty signal, we employ the difference between the level of OVX
and its 12-month moving average. We use this as our oil signal in Equation (4) and evaluate
the performance of the XTSMOM strategy. The third set of bars in Figure 3 plots the Sharpe
ratios of XTSMOM, TSMOM and buy & hold strategies based on this alternative oil
uncertainty signal. The XTSMOM strategy obtains a Sharpe ratio of 0.168 which is superior
to those of TSMOM (0.121) and buy & hold (0.062) strategies. Therefore, we can conclude
that the performance of XTSMOM strategies does not hinge on how we define the oil
uncertainty signal.
3.5.3. Transaction Costs and Short-Selling Constraints
26
The profitability of the TSMOM and XTSMOM strategies might be affected by transaction
costs. To address this concern, we calculate the net return of the buy & hold, TSMOM and
XTSMOM strategies as follows,
𝑒,𝑖
𝑃
𝑟𝑡+1 = ∑𝑁 𝑁
𝑖=1 𝑤𝑖,𝑡 𝑟𝑡+1 − ∑𝑖=1 𝑇𝐶𝑖 ∙ |𝑤𝑖,𝑡 − 𝑤𝑖,𝑡−1+ |, (8)
where 𝑤𝑖,𝑡 is the weight assigned to the ith country index based on the buy & hold, TSMOM
𝑒,𝑖
or XTSMOM strategy at month t,21 𝑤𝑖,𝑡 + ≡ 𝑤𝑖,𝑡 ∙ (1 + 𝑟𝑡+1 ) is the actual portfolio weight right
𝑒,𝑖
before the next rebalancing at 𝑡 + 1, and 𝑟𝑡+1 is the monthly return of the ith country index
from month 𝑡 to month 𝑡 + 1. 𝑇𝐶𝑖 is the transaction costs for the ith country index. We follow
the transaction costs documented by Angelidis and Tessaromatis (2017), which we also report
in Appendix A. These transaction costs are based on the half-trading spread of BlackRock’s
ETFs and Global X’s ETFs, which track the MSCI stock market index for each country.22 The
results are reported in Figure 3. As expected, the performance of the XTSMOM strategy is
slightly weaker after the transaction costs are accounted for, with a monthly XTSMOM Sharpe
ratio of 0.126. However, it is still superior to that of the TSMOM (0.069) and buy & hold (0.027)
strategies.23
21
The weights of the buy & hold are 𝑤𝑖,𝑡 = 1⁄𝑁 , those of the TSMOM are 𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛(𝑟𝑡𝑇𝑆𝑀𝑂𝑀,𝑖 )⁄𝑁 , and those
of the XTSMOM are 𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛(𝑟𝑡𝑋𝑇𝑆𝑀𝑂𝑀,𝑖 )⁄𝑁.
22
For countries for which transaction costs are not available, we take a conservative approach and use the highest
half-trading spread (i.e., 0.43%) reported in Angelidis and Tessaromatis (2017).
23
We also calculated the hypothetical breakeven transaction cost that would make our strategy unprofitable. We
find that the hypothetical half spread, averaged across all the sample countries, is 0.72%. This is substantially
higher than the maximum half spread of the ETFs in our sample (0.43%).
27
Currently, our strategies allow for short positions in international markets. In practice,
short selling may be prohibited in some markets. We therefore replicate the baseline results
with short-selling restrictions, i.e., we only allow long positions in our portfolios and invest in
the risk-free rate when our strategies suggest going short. We obtained an even better
XTSMOM performance with a Sharpe ratio of 0.201 (compared to 0.175 that is reported in
Panel A of Table 4) for all the countries, and this is still superior to the TSMOM (0.163) and
buy & hold (0.069) benchmarks that also do not allow short-selling. Thus, our main finding is
not driven by the possibility of shorting global equity markets.
3.5.4. Stock Market Excess Returns in Local Currency
To ensure that our results are not driven by the dollar effect, we test if our results are robust to
the currency used for calculating the stock market returns. Specifically, we consider a US
investor who invests locally, i.e., we conduct all our analysis using equity returns in local
currencies (in excess of the local risk-free rate). Stock market data and short-term interest rates
are obtained from Refinitiv Datastream. Figure 3 shows that the choice of the currency does
not affect the performance of the XTSMOM strategy. The monthly XTSMOM Sharpe ratio is
0.176, whereas those of the TSMOM and buy & hold strategies are 0.012 and 0.050,
respectively. Therefore, our results hold regardless of whether we use excess returns in US
dollars or local currency.
28
3.5.5. Scaled XTSMOM Returns
So far, our reported results are based on the unscaled XTSMOM excess returns. PSV (2020)
scale up all the portfolio weights of the cross-asset strategy so that, for each month, the amount
of capital allocated to the active positions is the same for both the XTSMOM and TSMOM
strategies. Thus, we scale the XTSMOM excess returns following PSV (2020). Figure 3 further
shows that the scaled XTSMOM offers a monthly Sharpe ratio of 0.170, which outperforms
those of the TSMOM (0.114) and buy & hold (0.069) strategies.
3.5.6. Orthogonalized XTSMOM Strategy
Robe and Wallen (2016) find that the OVX is closely linked to the VIX. In this robustness test,
we orthogonalize the OVX against the VIX. We follow the Modified Gram-Smith process,
which is a method commonly used in mathematics for orthogonalizing a set of vectors. We use
crude oil uncertainty signal based on the sign of the orthogonalized OVX changes
(ΔOVX_orthog) in Equation (4), instead of OVX changes (ΔOVX). We then recalculate the
XTSMOM returns.
The performance of XTSMOM is slightly weaker once the OVX has been
orthogonalized against the VIX. For instance, the XTSMOM Sharpe ratio with the
orthogonalized OVX changes is 0.138 (compared to 0.175 with just the OVX changes). This
implies that part of the predictability in the OVX signal comes from the VIX. However,
XTSMOM still outperforms the benchmarks, suggesting that the signal provided by OVX is
29
also useful for predicting stock market returns. We postulate this is because OVX captures
changes in the expectations of oil market participants such as the fundamentals in the oil
demand and supply, as well as fears and changes in oil market participants’ risk aversion. These
factors are not captured by the VIX. Our results show that these oil-specific factors contain
predictive information about the stock markets beyond the information contained in the VIX.
3.5.7. Subsample Analysis
When it was first introduced, the USO was structured to take positions in the front-month WTI
light sweet crude oil futures. Over the years, this holding policy changed as follows: (1) in
January 2009, the USO started using the front month through WTI futures and over-the-counter
positions; (2) in July 2013, the USO returned to using only the front-month WTI futures
contract; (3) in April 2020, USO started holding positions in various WTI futures maturities in
addition to the front-month WTI futures contract. 24 These policy changes might affect our
results. To assess whether these changes affect our results, we split our sample into four periods:
(1) May 2007 to December 2008, (2) January 2009 to June 2013, (3) July 2013 to March 2020,
and (4) April 2020 to August 2021. Then, we evaluate the performance of each trading strategy.
The results of this subsample analysis are reported in Appendix F.
In general, we observe that the XTSMOM strategy outperforms TSMOM in all
subperiods except the last one (April 2020 - August 2021). We note that this subperiod is the
24
We confirmed this using the (end-of-month) USO holdings data from Morningstar.
30
shortest (17 months) and starts with the erratic month of April 2020, when the crude oil futures
price fell below zero for the first time since its inception in March 1983. Surprisingly, during
this last subperiod, buy & hold offered much higher returns than the other two strategies in all
subperiods, and the returns on TSMOM and XTSMOM are nearly zero. Nevertheless, more
data are needed in the future to examine the impact on the results of the policy change in April
2020.25
Next, we assess whether the XTSMOM performance is affected by crisis periods. We
therefore consider periods of NBER (National Bureau of Economic Research) recessions and
expansions. Brusa, Savor, and Wilson (2020) show that US macroeconomic policy has a larger
effect on foreign country stock markets than local macroeconomic policy. As such, we use the
NBER cycle to identify crises in our sample countries. The middle two columns in Appendix
F show that XTSMOM offers a higher Sharpe ratio than TSMOM during expansions and,
especially, recessions. As expected, the buy & hold strategy performs well during expansions
and poorly during recessions. Therefore, we conclude that financial crises do not affect the
outperformance of XTSMOM relative to TSMOM. Finally, we split the sample into two equal
subsamples to ensure that our results hold across different periods in our sample. The first
25
The COVID-19 pandemic might also affect OVX change signals since the first quarter of 2020: first, a large
drop in demand without a decrease in production, then, a substantial decrease in production, and, finally, strong
recovery in demand without a similar increase in production (Today in Energy, 2021). An alternative reason for
the results of the last subperiod is that the USO invests in many contracts along the curve. Thus, it is related less
to current oil price expectations and more to the average of current and future expectations. All this warrants
further research.
31
subperiod is from June 2007 to July 2014 and the second is from August 2014 – August 2021.
The last two columns in Appendix F show that in both subperiods, the Sharpe ratio of
XTSMOM strategy is higher compared to the benchmarks.
4. Additional Results
4.1. The XTSMOM Smile
Moskowitz et al. (2012) demonstrate that, when plotted against stock market returns, the
returns of the TSMOM strategy take the shape of a smile, suggesting that the time series
momentum strategy works better under extreme market conditions. We examine whether the
same is true of the XTSMOM strategy when we compare monthly excess returns of the
TSMOM and XTSMOM portfolios against the monthly developed stock market excess returns
from Kenneth French’s website.
In Figure 4, we observe that both the TSMOM and XTSMOM strategies take a smile
shape, though they are not perfectly symmetrical. The shape of the XTSMOM smile is more
pronounced in both the positive and negative return domains. These results suggest that, from
a portfolio diversification perspective, the XTSMOM portfolio is valuable because of the
slightly higher returns it offers during periods when the market return is negative. Similarly,
the XTSMOM portfolio also offers higher returns during periods when the market return is
32
positive.
4.2. The XTSMOM and the Economy
Finally, we explore the possible link between the XTSMOM strategy and the real economy by
investigating future changes in economic variables, such as industrial production,
unemployment, and inflation, under various TSMOM and XTSMOM regimes. To do so, we
report the average next-12-month changes in economic indicators under momentum regimes
with positive/negative stock returns, positive/negative OVX changes, and a combination of
both.
Table 9 reports the average values of the macroeconomic variables in basis points under
various momentum regimes. Panel A shows that positive stock returns and negative OVX
changes are associated with better future economic prospects—that is, higher industrial
production, a lower unemployment rate, and lower inflation. Therefore, the TSMOM seems to
be related to future economic outcomes.
Panel B reports the averages of the macroeconomic variables across different
XTSMOM regimes. Periods of positive stock returns and negative OVX changes are associated
with the highest industrial production (IP) index changes and the lowest unemployment rate.
In addition, these periods are associated with declining inflation. Therefore, the long
XTSMOM portfolio predicts good economic times. Periods of negative stock returns and
positive OVX changes are associated with the lowest IP changes, the largest increase in the
33
unemployment rate, and the highest inflation rate. Therefore, the short XTSMOM portfolio
predicts hard economic times.
These results are consistent with the literature showing that uncertainty in crude oil
prices has a significant impact on the global economy and stock market returns (e.g., Guo and
Kliesen, 2005; Jo, 2014; Kwon, 2020; Gao et al., 2022). For instance, Gao et al. (2022) find
that option-implied oil volatility is a strong negative predictor of economic growth beyond the
standard financial, macroeconomic, and policy uncertainty measures. Therefore, we conclude
that the XTSMOM is superior to the TSMOM strategy in predicting economic cycles.
5. Conclusions
We document a cross-asset time-series momentum (XTSMOM) strategy in crude oil volatility
and international stock markets. Using a sample of 59 stock markets in 44 oil-importing and
15 oil-exporting countries, we show that past stock excess returns are positive predictors and
past crude oil implied volatility index (OVX) changes are negative predictors of future stock
market excess returns. This XTSMOM strategy outperforms the TSMOM in terms of larger
mean excess returns, lower standard deviations, and higher Sharpe ratios. The predictive power
of oil price uncertainty for the stock markets can be explained by the funding constraints of
financial intermediaries. In addition, we show that the XTSMOM contains information about
future changes in real economic conditions. Specifically, the long (short) XTSMOM
portfolio—namely, positive (negative) stock returns and negative (positive) OVX changes —
34
can point to good (bad) economic times, with higher (lower) industrial production, declining
(increasing) unemployment rates, and less (more) inflation. The XTSMOM gives a better
indication of future economic activity than a single-asset TSMOM strategy.
Our study is the closest to PSV (2020) who document an XTSMOM strategy in bond
and stock markets. We contribute to the literature by constructing an XTSMOM strategy
between option-implied crude oil volatility and stock returns. We also contribute to research
on the impacts of oil price uncertainty on stock market returns (e.g., Christoffersen and Pan,
2018; Kwon, 2020; Gao et al., 2022).
35
References
Amihud, Y., 2002. Illiquidity and stock returns: Cross-section and time-series effects.
J. Financial Mark. 5 (1), 31-56.
Angelidis, T., Tessaromatis, N., 2017. Global equity country allocation: an application of factor
investing. Financial Anal. J. 73 (4), 55–73.
Asness, C. S., Moskowitz, T. J., Pedersen, L. H., 2013. Value and momentum everywhere. J.
Finance 68 (3), 929–985.
Barsky, R. B., Kilian, L., 2004. Oil and the macroeconomy since the 1970s. J. Econ. Perspect.,
18, 115-134.
Bekaert, G., Hoerova, M., Duca, M.L., 2013. Risk, uncertainty and monetary policy.
J. Monet. Econ. 60 (7), 771-788.
Bekaert, G., Hoerova, M., 2014. The VIX, the variance premium and stock market volatility. J.
Econom. 183, 181-192.
Bernanke, B., 1983. Irreversibility, uncertainty, and cyclical investment. Quarterly J. of Econ.
98, 85-106.
Brusa, F., Savor, P., Wilson, M., 2020. One central bank to rule them all. Rev. Financ 24 (2),
263-304.
Büyükşahin, B., Haigh, M.S., Robe, M.A., 2010. Commodities and equities: Ever a “market of
one”? J. Altern. Invest. 12, 76-95.
36
Büyükşahin, B., Robe, M.A., 2014. Speculators, commodities and cross-market linkages. J.
Money Credit Bank 42, 38-70.
Carhart, M. M., 1997. On persistence in mutual fund performance. J. Finance 52 (1), 57–82.
Cheng, I.-H., Xiong, W., 2014. Financialization of commodity markets. Annu. Rev. Financial
Econ. 6, 419-441.
Cheng, I.-H., Kirilenko, A., Xiong, W., 2015. Convective risk flows in commodity futures
markets. Rev. Financ 19, 1733-1781.
Christoffersen, P., Pan, X., 2018. Oil volatility risk and expected stock returns. J. Bank. Finance
95, 5–26.
Coval, J. D., Shumway, T., 2001. Expected option returns. J. Finance 56 (3), 983–1009.
Diaz, E. M., Molero, J. C., de Gracia, F. P., 2016. Oil price volatility and stock returns in the
G7 economies. Energy Econ. 54, 417–430.
Diebold, F.X., Liu, L, Yilmaz, K., 2017. Commodity connectedness. NBER Working Paper
23685.
Elder, J., Serletis, A., 2010. Oil price uncertainty. J. Money. Credit. Bank 42, 1137-1159.
Fama, E. F., French, K. R., 1993. Common risk factors in the returns on stocks and bonds. J.
Financ. Econ. 33 (1), 3–56.
Frazzini, A., & Pedersen, L. H. (2014). Betting against beta. J. Financ. Econ. 111 (1), 1-25.
Gao, L., Hitzemann, S., Shaliastovich, I., Xu, L., 2022. Oil volatility risk. J. Financ. Econ. 144
(2), 456-491.
37
Gorton, G. B., Hayashi, F., Rouwenhorst, K. G., 2013. The fundamentals of commodity futures
returns. Rev. Financ, 17, 35-105
Goyal, A., Jegadeesh, N., 2018. Cross-sectional and time-series tests of return predictability:
what is the difference? Rev. Financ. Stud. 31 (5), 1784–1824.
Guo, H., Kliesen, K., 2005. Oil price volatility and US macroeconomic activity. Fed. Reserve
Bank St. Louis Rev. 87, 669–684.
Hamilton, J.D., 2009. Causes and consequences of the oil shock of 2007-08. NBER Working
Paper, 15002.
Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: Implications for
stock market efficiency. J. Finance 48 (1), 65–91.
Jo, S., 2014. The effects of oil price uncertainty on global real economic activity. J. Money.
Credit. Bank 46(6), 1113-1135.
Jones, C.M., Kaul, G., 1996. Oil and the stock markets. J. Finance 51(2), 463-491.
Jostova, G., Nikolova, S., Philipov, A., Stahel, C. W., 2013. Momentum in corporate bond
returns. Rev. Financ. Stud. 26(7), 1649–1693.
Kellogg, R., 2014. The effect of uncertainty on investment: evidence from Texas oil drilling.
Am. Econ. Rev. 104(6), 1698-1734.
Kilian, L. (2009). Not all oil price shocks are alike: Disentangling demand and supply shocks
in the crude oil market. Am. Econ. Rev. 99(3), 1053-1069.
Kilian, L., C. Park (2009). The impact of oil price shocks on the U.S. stock market. Int. Econ.
38
Rev. 50(4), 1267-1287.
Kilian, L., Vega, C., 2011. Do energy prices respond to US macroeconomic news? A test of the
hypothesis of predetermined energy prices. Rev. Econ. Stat. 93, 660-671.
Kim, A. Y., Tse, Y., Wald, J. K., 2016. Time series momentum and volatility scaling. J. Financial
Mark. 30, 103–124.
Kling, J.L., 1985. Oil price shocks and stock market behavior. J. Portf. Manag. 12(1), 34-39.
Kocaaslan, O.K., 2019. Oil price uncertainty and unemployment. Energy Econ. 81, 577-583.
Kollias, C., Kyrtsou, C., Papadamou, S. 2013. The effects of terrorism on the oil price stock
index relationship. Energy Econ. 40, 743-752.
Kwon, D., 2020. The impacts of oil price shocks and United States economic uncertainty on
global stock markets. Int. J. Finance Econ., 1-13.
Menkhoff, L., Sarno, L., Schmeling, M., Schrimpf, A., 2012. Currency momentum strategies. J.
Financ. Econ. 106 (3), 660–684.
Miffre, J., Rallis, G., 2007. Momentum strategies in commodity futures markets. J. Bank.
Finance 31, 1863–1886.
Moskowitz, T. J., Ooi, Y. H., Pedersen, L. H., 2012. Time series momentum. J. Financ.
Econ. 104 (2), 228–250.
Nonejad, N., 2020. Crude oil price volatility and short-term predictability of the real US GDP
growth rate. Econ. Let. 186, 108527.
Opdyke, J. D. J., 2007. Comparing Sharpe ratios: So where are the p-values? J. Asset Manag. 8
39
(5), 308–336.
Pindyck, R.S., 1991. Irreversibility, uncertainty, and investment. J. Econ. Lit. 29, 1110-1148.
Pitkäjärvi, A., Suominen, M., Vaittinen, L., 2020. Cross-asset signals and time series
momentum. J. Financ. Econ. 136 (1), 63–85.
Rapach, D. E., Strauss, J. K., Zhou, G., 2013. International stock return predictability: what is
the role of the United States? J. Finance, 68, 1633-1662.
Robe, M. A., Wallen, J., 2016. Fundamentals, derivatives market information and oil price
volatility. J. Futures Mark. 36, 317–344.
Rouwenhorst, G. K., 1998. International momentum strategies. J. Finance 53, 267–284.
Tang, K., Xiong, W., 2012. Index investment and financialization of commodities. Financial
Anal. J. 68, 54-74.
Today in Energy, 2021. Independent Statistics & Analysis. US Energy Information
Administration, January 5.
Wang, T., 2021. Local banks and the effects of oil price shocks. J. Bank. Finance 125, 106069.
Wang, Y., Wu, C., Yang, L., 2013. Oil price shocks and stock market activities: Evidence from
oil-importing and oil-exporting countries. J. Comp. Econ., 41, 1220-1239.
Working, H., 1933. Price relations between July and September wheat futures at Chicago since
1885. Wheat Studies, 9, 187-240.
40
Appendix A. List of countries in the sample
Panel A: Oil-importing countries Panel B: Oil-exporting countries
No. Country Return index TC No. Country Return index TC
1 Australia MSAUST$ 0.0215% 1 Argentina MSARGT$ 0.4312%
2 Austria MSASTR$ 0.1306% 2 Brazil MSBRAZ$ 0.0149%
3 Bangladesh MSBNGS$ 0.4312%* 3 Canada MSCNDA$ 0.0182%
4 Belgium MSBELG$ 0.0537% 4 Colombia MSCOLM$ 0.4312%
5 Bulgaria MSBLGN$ 0.4312%* 5 Denmark MSDNMK$ 0.0800%
6 Chile MSCHIL$ 0.0621% 6 Egypt MSEGYT$ 0.4312%*
7 China MSCHIN$ 0.0248% 7 Estonia MSESTN$ 0.4312%*
8 Croatia MSCROA$ 0.4312%* 8 Malaysia MSMALF$ 0.0380%
9 Czech Republic MSCZCH$ 0.4312%* 9 Mexico MSMEXF$ 0.0122%
10 Finland MSFIND$ 0.1226% 10 Norway MSNWAY$ 0.1201%
11 France MSFRNC$ 0.0195% 11 Qatar MSQATA$ 0.4312%*
12 Germany MSGERM$ 0.0168% 12 Russia MSRUSS$ 0.0681%
13 Greece MSGREE$ 0.0746% 13 Saudi Arabia MSSAUD$ 0.4312%*
14 Hong Kong MSHGKG$ 0.0227% 14 UAE MSUAEI$ 0.4312%*
15 Hungary MSHUNG$ 0.4312%* 15 Vietnam MSVIET$ 0.4312%*
16 India MSINDI$ 0.0159%
17 Indonesia MSINDF$ 0.0421%
18 Ireland MSEIRE$ 0.1620%
19 Israel MSISRL$ 0.1682%
20 Italy MSITAL$ 0.0336%
21 Jamaica MSJMCA$ 0.4312%*
22 Japan MSJPAN$ 0.0434%
23 Lithuania MSLITH$ 0.4312%*
24 Morocco MSMORC$ 0.4312%*
25 Netherlands MSNETH$ 0.0339%
26 New Zealand MSNZEA$ 0.1542%
27 Pakistan MSPAKI$ 0.4312%*
28 Peru MSPERU$ 0.1269%
29 Philippines MSPHLF$ 0.0614%
30 Poland MSPLND$ 0.4312%*
31 Portugal MSPORD$ 0.1689%
32 Romania MSROMN$ 0.4312%*
33 Serbia MSSERB$ 0.4312%*
34 Singapore MSSING$ 0.0388%
35 Slovenia MSSLVN$ 0.4312%*
36 South Africa MSSARF$ 0.0592%
37 South Korea MSKORE$ 0.0087%
38 Spain MSSPAN$ 0.0152%
39 Sweden MSSWDN$ 0.0359%
40 Taiwan MSTAIW$ 0.0315%
41 Thailand MSTHAF$ 0.0673%
42 Turkey MSTURK$ 0.0657%
43 UK MSUTDK$ 0.0268%
44 US MSUSAM$ 0.0042%
This table reports the list of countries in the data sample and the Refinitiv Datastream ticker symbols for the stock market
index for 44 oil-importing (Panel A) and 15 oil-exporting countries (Panel B). TC are the transaction costs from Angelidis and
Tessaromatis (2017). The sample period is from May 2007 to August 2021. * indicates the highest available transaction cost.
41
Appendix B. Time-series analysis: oil-importing countries
Dependent variables: 𝑟𝑡𝑒,𝑖 𝛼 𝛽𝑒,𝑖 𝛽𝑂𝑉𝑋,𝑖 𝐴𝑑𝑗 𝑅2 (%)
Australia 0.004 [0.74] 0.082 [1.00] -0.012** [–2.29] 3.14
Austria 0.000 [0.04] 0.091 [0.81] -0.011 [-1.45] 1.55
Bangladesh 0.001 [0.26] 0.048 [0.86] -0.010** [-2.25] 1.09
Belgium 0.001 [0.09] 0.119 [1.00] -0.010** [-2.26] 3.47
Bulgaria -0.006 [-0.84] 0.293** [2.31] -0.006 [-0.76] 8.67
Chile 0.001 [0.20] -0.081 [-0.77] -0.006 [-1.13] -0.22
China 0.006 [0.99] 0.090 [1.06] -0.006 [-1.26] -0.54
Croatia 0.000 [0.05] -0.025 [-0.22] -0.014*** [-2.97] 3.16
Czech Republic 0.003 [0.42] 0.068 [0.92] -0.013** [-2.37] 2.66
Finland 0.004 [0.68] 0.050 [0.66] -0.006 [-1.17] 0.01
France 0.004 [0.78] -0.014 [-0.20] -0.008 [-1.61] 0.29
Germany 0.004 [0.77] -0.008 [-0.11] -0.010** [-1.98] 0.52
Greece -0.012 [-1.24] 0.074 [0.90] -0.005 [-0.51] -0.33
Hong Kong 0.006 [1.17] 0.037 [0.34] -0.007* [-1.75] 0.53
Hungary 0.003 [0.43] 0.065 [0.70] -0.015** [-2.11] 2.04
India 0.007 [1.08] 0.005 [0.07] -0.007 [-1.28] -0.44
Indonesia 0.006 [0.83] 0.129 [1.37] -0.007 [-1.07] 1.62
Ireland -0.001 [-0.11] 0.089 [0.76] -0.005 [-0.92] 0.57
Israel 0.002 [0.49] -0.010 [-0.14] -0.013*** [-3.14] 4.26
Italy 0.001 [0.12] 0.036 [0.51] -0.006 [-1.09] -0.20
Jamaica 0.011* [1.87] 0.055 [0.69] -0.001 [-0.15] -1.03
Japan 0.002 [0.68] 0.004 [0.05] -0.011*** [-2.86] 4.27
Lithuania 0.006 [1.00] 0.146 [0.94] -0.003 [-0.38] 1.30
Morocco 0.001 [0.35] 0.024 [0.20] -0.009** [-1.97] 2.60
Netherlands 0.007 [1.31] 0.018 [0.19] -0.007 [-1.60] 0.18
New Zealand 0.005 [0.96] 0.054 [0.79] -0.012** [-2.33] 3.09
Pakistan -0.002 [-0.27] 0.009 [0.11] -0.019*** [-2.75] 3.68
Peru 0.005 [0.66] -0.058 [-0.66] -0.009 [-1.46] -0.12
Philippines 0.005 [0.95] -0.002 [-0.04] -0.006 [-1.23] -0.44
Poland 0.001 [0.11] 0.025 [0.37] -0.010 [-1.56] 0.17
Portugal -0.001 [-0.25] 0.049 [0.73] -0.005 [-1.02] -0.15
Romania 0.005 [0.64] 0.078 [1.00] -0.019** [-2.31] 3.11
Serbia -0.004 [-0.51] 0.194** [2.10] -0.017** [-2.40] 5.73
Singapore 0.003 [0.62] 0.008 [0.08] -0.012*** [-2.87] 2.03
Slovenia 0.000 [0.06] 0.085 [0.79] -0.012** [-2.58] 3.33
South Africa 0.004 [0.73] -0.067 [-1.06] -0.010* [-1.87] 0.57
South Korea 0.006 [0.92] -0.013 [-0.19] -0.011** [-2.08] 0.69
Spain 0.002 [0.29] 0.000 [0.00] -0.007 [-1.10] -0.41
Sweden 0.006 [1.03] 0.039 [0.43] -0.010* [-1.84] 1.20
Taiwan 0.008 [1.45] 0.024 [0.23] -0.016*** [-3.52] 5.35
Thailand 0.007 [1.22] 0.061 [0.66] -0.005 [-0.80] -0.25
Turkey 0.001 [0.15] 0.031 [0.42] -0.006 [-0.80] -0.77
UK 0.002 [0.34] 0.068 [0.71] -0.006 [-1.51] 0.96
US 0.009** [2.06] -0.071 [-0.85] -0.009** [-2.19] 1.14
This table reports the predictability of cross-asset time series using the signs of one-month lagged equity market
returns and OVX changes per country—that is, 𝑟𝑡𝑒,𝑖 = 𝛼 + 𝛽 𝑒,𝑖 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) + 𝛽 𝑂𝑉𝑋,𝑖 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−1 ) + 𝜀𝑡𝑖 for
i=1, …, N, and N is the total number of countries in our sample, for the subsample of oil-importing countries.
Newey-West corrected t-statistics are reported in square brackets. *, **, and *** represent significance at the 10%,
5%, and 1% levels, respectively. The sample period is from May 2007 to August 2021.
42
Appendix C. Time-series analysis: oil-exporting countries
Dependent variables: 𝑟𝑡𝑒,𝑖 𝛼 𝛽𝑒,𝑖 𝛽𝑂𝑉𝑋,𝑖 𝐴𝑑𝑗 𝑅2 (%)

Argentina 0.006 [0.67] 0.088 [0.92] -0.013 [-1.21] 0.96
Brazil 0.004 [0.47] 0.080 [1.08] -0.012 [-1.62] 1.53
Canada 0.004 [0.80] 0.003 [0.03] -0.010** [-2.45] 1.32
Colombia 0.004 [0.60] -0.029 [-0.41] -0.009 [-1.52] -0.27
Denmark 0.007 [1.37] 0.169 [1.38] -0.008* [-1.87] 4.77
Egypt 0.001 [0.09] -0.019 [-0.22] -0.014** [-2.02] 1.24
Estonia 0.003 [0.39] -0.038 [-0.33] -0.014** [-2.04] 1.18
Malaysia 0.001 [0.35] 0.129* [1.91] -0.008** [-2.36] 4.29
Mexico 0.002 [0.38] -0.007 [-0.11] -0.011* [-1.68] 0.58
Norway 0.002 [0.34] 0.027 [0.22] -0.015*** [-2.63] 2.58
Qatar 0.006 [1.02] 0.070 [0.79] -0.006 [-1.26] 0.32
Russia 0.004 [0.44] 0.104 [0.89] -0.015* [-1.94] 3.39
Saudi Arabia 0.006 [0.94] 0.046 [0.45] 0.000 [0.06] -2.29
UAE 0.003 [0.39] 0.141 [0.91] -0.009 [-1.61] 2.90
Vietnam 0.002 [0.38] 0.142** [2.20] -0.003 [-0.35] 1.13
This table reports the predictability of cross-asset time series using the signs of one-month lagged equity market
returns and OVX changes per country—that is, 𝑟𝑡𝑒,𝑖 = 𝛼 + 𝛽 𝑒,𝑖 𝑠𝑖𝑔𝑛(𝑟𝑡−1
𝑒,𝑖
) + 𝛽 𝑂𝑉𝑋,𝑖 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋𝑡−1 ) + 𝜀𝑡𝑖 for
i=1, …, N, and N is the total number of countries in our sample, for the subsample of oil-exporting countries.
43
Appendix D. Future value of $1 invested in the trading strategies with alternative initial
investment dates
Panel A: May 2007-June 2014
Panel B: June 2014-August 2021
The figure plots the future value of $1 invested in the buy & hold, TSMOM, and XTSMOM for all the countries
with alternative initial investment dates. The strategies include the US one-month risk-free rate. The sample period
is from May 2007 to June 2014 in Panel A and from June 2014 to August 2021 in Panel B.
44
Appendix E. Spanning tests
Alpha (%) TSMOM XTSMOM Adj 𝑅2 (%)

Panel A: All countries
XTSMOM 0.360** 0.834*** 69.66
[2.05] [9.52]
TSMOM -0.145 0.838*** 69.66
[-1.00] [14.98]
Panel B: Oil-importing countries
XTSMOM 0.366** 0.817*** 67.90
[1.99] [8.67]
TSMOM -0.154 0.833*** 67.90
[-0.98] [14.12]
Panel C: Oil-exporting countries
XTSMOM 0.369** 0.831*** 73.10
[2.25] [12.64]
TSMOM -0.146 0.881*** 73.10
[-1.02] [23.53]
This table reports the results of regressing the monthly returns of the XTSMOM and TSMOM portfolios on each
other. The results for all countries, oil-importing, and oil-exporting countries are presented in Panels A, B, and C,
respectively. Newey-West corrected t-statistics are reported in square brackets. ** and *** represent significance
at the 5% and 1% levels, respectively. The sample period is from May 2007 to August 2021.
45
Appendix F. Subsample analysis
USO holding policies US economic cycle Two halves

May 2007 - Jan 2009 - Jul 2013 - Apr 2020 - Jun 2007 - Aug 2014 -
Recession Expansion
Dec 2008 Jun 2013 Mar 2020 Aug 2021 Jul 2014 Aug 2021
#Obs 19 54 81 17 22 149 86 85
Buy & hold -0.355 0.188 0.018 0.676 -0.232 0.203 0.050 0.100
TSMOM 0.319 0.059 0.095 0.041 0.315 0.041 0.137 0.084
XTSMOM 0.368 0.151 0.184 -0.056 0.456 0.091 0.212 0.127

Buy & hold -0.369 0.175 0.023 0.649 -0.240 0.197 0.039 0.107
TSMOM 0.317 0.041 0.088 0.038 0.293 0.033 0.125 0.077
XTSMOM 0.354 0.135 0.186 -0.051 0.443 0.084 0.197 0.130

Buy & hold -0.312 0.223 0.006 0.717 -0.205 0.210 0.084 0.077
TSMOM 0.306 0.107 0.101 0.046 0.367 0.052 0.162 0.092
XTSMOM 0.396 0.191 0.169 -0.066 0.486 0.101 0.246 0.114
This table reports the number of month observations in each subsample (i.e., #Obs) and the Sharpe ratio of various trading strategies, such as buy & hold, TSMOM, and
XTSMOM for several subperiods based on (1) USO’s holding policies, (2) NBER recession vs. expansion periods and (3) splitting the sample in two halves. Statistics for all
countries, oil-importing, and oil-exporting countries are presented in Panels A, B, and C, respectively.
46
Table 1. Descriptive statistics for the OVX changes and stock market excess returns
Country Mean (%) t-stat Std. dev. (%) Skew. Kurt. AR(1)
OVX changes 0.051 [0.07] 12.181 3.834 43.457 -0.057
All countries 0.417*** [4.98] 7.978 -0.260 6.956 0.089
Oil-importing countries 0.392*** [4.05] 7.790 -0.246 7.029 0.087
Oil-exporting countries 0.494*** [2.73] 8.519 -0.295 6.667 0.093
This table reports the descriptive statistics such as the mean, standard deviation, skewness, kurtosis, and first-
order autocorrelation coefficient for the OVX changes and pooled stock market excess returns at monthly
frequency. Newey-West corrected t-statistics are reported in square brackets. *** represents statistical significance
at the 1% level. The sample period is from May 2007 to August 2021.
47
Table 2. Cross-asset time-series predictability
Constant 𝑠𝑖𝑔𝑛(𝑟 𝑒 ) 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋) 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋) 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝐶𝑉) 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃) 𝑆𝑀𝐵 𝐻𝑀𝐿 𝐴𝑑𝑗 𝑅2 (%)

0.004 0.005 0.33
[0.89] [1.52]
0.003 0.003 -0.010*** 1.79
[0.78] [0.94] [-3.59]
0.003 0.002 -0.010** -0.001 -0.102*** 0.154*** 3.30
[0.64] [0.50] [-2.11] [-0.26] [-2.90] [3.55]
0.003 0.001 -0.009* 0.002 -0.007* -0.104*** 0.153*** 4.02
[0.64] [0.29] [-1.90] [0.33] [-1.70] [-2.99] [3.55]

0.003 0.004 0.27
[0.83] [1.32]
0.003 0.002 -0.010*** 1.73
[0.72] [0.76] [-3.59]
0.003 0.002 -0.009** -0.001 -0.130*** 0.152*** 3.58
[0.61] [0.49] [-1.98] [-0.29] [-3.86] [3.39]
0.003 0.001 -0.008* 0.001 -0.007 -0.132*** 0.152*** 4.30
[0.60] [0.26] [-1.76] [0.21] [-1.62] [-3.92] [3.39]

0.004 0.006** 0.49
[1.02] [1.96]
0.004 0.004 -0.011*** 1.92
[0.92] [1.37] [-3.35]
0.004 0.001 -0.011** 0.000 0.026 0.165** 3.12
[0.79] [0.43] [-2.36] [-0.11] [0.33] [2.30]
0.004 0.001 -0.010** 0.004 -0.008* 0.022 0.164** 3.85
[0.83] [0.33] [-2.22] [0.77] [-1.83] [0.29] [2.31]
This table reports the predictability of single-asset time series using the sign of one-month lagged stock market
returns (first model), and the predictability of cross-asset time series using the signs of one-month lagged stock
market returns and OVX changes (second to fourth models). Panel A reports the results for all countries, while
Panels B and C report the result for the oil-importing and oil-exporting countries, respectively.
𝑠𝑖𝑔𝑛(𝑟 𝑒 ), 𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋), 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋), 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝐶𝑉) and 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃 ) are the sign of the stock excess
returns, OVX changes, VIX changes, changes in the VIX conditional variance, and changes in the VIX variance
premium, respectively. 𝑆𝑀𝐵 and 𝐻𝑀𝐿 are the one-month lagged size and book-to-market long-short portfolio
returns per country. White-corrected t-statistics are clustered by year-month and reported in square brackets. *,
**, and *** represent significance at the 10%, 5%, and 1% levels, respectively. The sample period is from May
2007 to August 2021.
48
Table 3. Returns by momentum regime
Panel A: TSMOM regime

Positive stock Negative stock Positive OVX Negative OVX
#Obs 5,345 4,525 4,406 5,352
Stock market return 0.84% -0.10% -0.80% 1.32%
Sharpe ratio 0.12 -0.01 -0.09 0.19
Panel B: XTSMOM regime

Positive stock, Negative stock, Positive stock, Negative stock,
positive OVX positive OVX negative OVX negative OVX
#Obs 1,883 2,520 3,402 1,947
Stock market return -0.24% -1.21% 1.38% 1.22%
Sharpe ratio -0.03 -0.12 0.22 0.15
This table reports the number of country-month combinations (i.e., #Obs), the average stock market excess return,
and corresponding Sharpe ratios under the TSMOM and XTSMOM regimes, presented in Panels A and B,
respectively. An asset is in a positive (negative) stock market regime in month t if the one-month lagged excess
return of the stock market is positive (negative). Likewise, an asset is in an XTSMOM regime of positive stock
returns and positive OVX changes if the one-month lagged excess return of the stock market and of OVX changes
are positive, and likewise for the other regimes. The sample period is from May 2007 to August 2021.
49
Table 4. Performance of various trading strategies
Mean (%) t-stat Std. dev. (%) Sharpe ratio Opdyke p-value Skewness Kurtosis

Buy & Hold 0.407 [0.76] 5.939 0.069 (0.194) -0.724 6.760
TSMOM 0.520 [1.63] 4.556 0.114* (0.053) 1.886 12.740
XTSMOM 0.794** [2.24] 4.545 0.175*** (0.004) 2.063 12.610

Buy & Hold 0.378 [0.71] 5.917 0.064 (0.420) -0.676 6.468
TSMOM 0.472 [1.55] 4.531 0.104 (0.150) 1.620 11.961
XTSMOM 0.751** [2.20] 4.488 0.167** (0.014) 1.911 12.016

Buy & Hold 0.498 [0.88] 6.223 0.080 (0.318) -0.794 7.049
TSMOM 0.669* [1.74] 5.079 0.132* (0.056) 1.992 11.545
XTSMOM 0.925** [2.29] 4.934 0.188*** (0.004) 2.202 12.466
This table reports the summary statistics, such as the mean, t-statistic, standard deviation, Sharpe ratio, Opdyke’s
(2007) Sharpe ratio p-value (whose null hypothesis is the Sharpe ratio equals zero), skewness, and kurtosis of the
excess returns of various trading strategies, such as buy & hold, TSMOM, and XTSMOM. Descriptive statistics
for all countries, oil-importing, and oil-exporting countries are presented in Panels A, B, and C, respectively.
50
Table 5. XTSMOM excess returns
Alpha (%) TSMOM MKT SMB HML UMD ∆𝑉𝐼𝑋 ∆𝑉𝐼𝑋_𝐶𝑉 ∆𝑉𝐼𝑋_𝑉𝑃 ∆𝐼𝐿𝐿𝐼𝑄 𝐴𝑑𝑗 𝑅2 (%)

0.413** 0.742*** 0.002 -0.045 -0.223* -0.234** 0.670* 3.788 72.62
[2.17] [8.56] [0.02] [-0.42] [-1.94] [-2.01] [1.79] [1.55]
0.449** 0.697*** 0.008 -0.020 -0.233** -0.251** 0.121 0.324** 3.055 73.90
[2.33] [7.73] [0.13] [-0.18] [-2.11] [-2.18] [1.40] [2.21] [1.43]

0.410** 0.725*** 0.004 -0.049 -0.233** -0.236** 0.690* 3.937* 71.16
[2.06] [7.66] [0.06] [-0.45] [-1.99] [-1.96] [1.67] [1.65]
0.444** 0.677*** 0.010 -0.024 -0.241** -0.254** 0.124 0.336** 3.154 72.45
[2.21] [6.94] [0.15] [-0.21] [-2.13] [-2.14] [1.49] [2.22] [1.53]

0.444** 0.748*** -0.009 -0.041 -0.197* -0.236** 0.778** 3.434 75.67
[2.54] [13.33] [-0.15] [-0.35] [-1.78] [-2.23] [2.53] [1.37]
0.485*** 0.718*** 0.000 -0.010 -0.213** -0.251** 0.138 0.321** 2.683 77.11
[2.77] [11.99] [-0.01] [-0.08] [-2.00] [-2.44] [1.29] [2.56] [1.16]
This table reports regression results from Equation (5) between the excess returns of the XTSMOM on the excess returns of the TSMOM and several standard asset pricing
factors {MKT, SMB, HML, UMD} for developed countries, and VIX returns and changes in its components (ΔVIX_CV and ΔVIX_VP) and ΔILLIQ. The results of the full
sample, subsample of oil-importing countries, and subsample of oil-exporting countries are presented in Panels A, B, and C, respectively. Newey-West corrected t-statistics are
reported in square brackets. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. The sample period is from May 2007 to August 2021.
51
Table 6. Excess returns of the XTSMOM everywhere
Alpha (%) TSMOM MSCI_World VAL_Everywhere MOM_Everywhere Δusd 𝐴𝑑𝑗 𝑅2 (%)

0.463*** 0.794*** -0.085 -0.287 -0.095* -0.137 70.24
[2.60] [10.00] [-0.51] [-1.42] [-1.85] [-1.36]

0.467** 0.775*** -0.078 -0.279 -0.097* -0.128 68.41
[2.48] [8.91] [-0.44] [-1.33] [-1.86] [-1.20]

0.500*** 0.792*** -0.109 -0.335* -0.124** -0.181* 74.13
[3.06] [13.97] [-0.66] [-1.76] [-2.03] [-1.89]
This table reports regression results from Equation (6) between the excess returns of the XTSMOM on the excess returns of the TSMOM, the MSCI World index, the Asness
et al. (2013) value, the momentum everywhere factors, and the natural log changes in the USD index. The results for all countries, oil-importing, and oil-exporting countries
are presented in Panels A, B, and C, respectively. Newey-West corrected t-statistics are reported in square brackets. *, **, and *** represent significance at the 10%, 5%, and
1% levels, respectively. The sample period is from May 2007 to August 2021.
52
Table 7. Changes in oil uncertainty and funding constraints of financial intermediaries
𝑇𝐸𝐷𝑡 𝐶𝑟𝑒𝑑𝑖𝑡𝑡 𝐵𝐴𝐵𝑡 𝐼𝐵𝐼𝑡
Constant 0.427*** [5.89] 2.739*** [20.78] 0.006*** [3.54] 0.001 [0.08]

𝛥𝑂𝑉𝑋𝑡−1 0.005 [1.51] 0.006 [1.16] -0.001*** [-3.59] -0.001* [-1.65]
Adj-R2 (%) 1.41 0.23 7.57 0.87
This table reports regression coefficients in the predictive regression from Equation (7) between the changes in
OVX on various funding constraint variables. These variables include the TED spread (𝑇𝐸𝐷), the Credit spread
(𝐶𝑟𝑒𝑑𝑖𝑡), the betting-against-beta factor (𝐵𝐴𝐵) and the International Bank Index returns (𝐼𝐵𝐼). Newey-West
corrected t-statistics are reported in square brackets. * and *** represent significance at the 10% and 1% levels,
respectively. The sample period is from May 2007 to August 2021.
53
Table 8. The impact of oil supply and demand shocks on XTSMOM returns
Alpha (%) Supply Agg. Demand Oil-specific Demand 𝐴𝑑𝑗 𝑅2 (%)
0.828*** -0.007* -0.002 -0.007* 2.65
[2.54] [-1.87] [-0.45] [-1.75]
0.783*** -0.006* -0.002 -0.006* 2.11
[2.49] [-1.80] [-0.44] [-1.67]
0.967*** -0.008** -0.003 -0.008* 3.79
[2.60] [-1.96] [-0.48] [-1.90]
This table reports regression results from Equation (8) between the excess returns of the XTSMOM on the crude
oil supply, aggregate demand, and oil-specific demand shocks from Kilian’s (2009) SVAR. The results for all
countries, oil-importing, and oil-exporting countries are presented in Panels A, B, and C, respectively. Newey-
West corrected t-statistics are reported in square brackets. *, **, and *** represent significance at the 10%, 5%,
and 1% levels, respectively. The sample period is from May 2007 to August 2021.
54
Table 9. XTSMOM and the economy
Panel A: TSMOM regime

Positive stock Negative stock Positive OVX Negative OVX
Average #Obs Average #Obs Average #Obs Average #Obs
IP (bps) 17.197 4,405 2.211 3,810 -1.192 3,755 19.766 4,547
Unemployment (bps) -1.448 4,136 24.119 3,549 26.495 3,473 -2.419 4,211
Inflation (bps) -1.776 4,855 -0.48 4,228 -0.139 4,104 -1.675 4,979
Panel B: XTSMOM regime

Positive stock, Positive stock, Negative stock, Negative stock,
positive OVX negative OVX positive OVX negative OVX
Average #Obs Average #Obs Average #Obs Average #Obs
IP (bps) 2.537 1,553 25.249 2,841 -3.968 2,134 10.560 1,633
Unemployment (bps) 7.301 1,468 -6.118 2,668 40.541 2,006 3.976 1,543
Inflation (bps) -4.05 1,713 -0.304 3,142 2.664 2,391 -4.025 1,837
This table reports the number of country-month combinations (i.e., #Obs), the average next-12-month changes in industrial production (i.e., IP), the unemployment and inflation
in basis points under the TSMOM and XTSMOM regimes, presented in Panels A and B, respectively. An asset is in a positive (negative) stock regime in month t if the one-
month lagged excess return of the stock market is positive (negative). Likewise, an asset is in an XTSMOM regime of positive stock returns and positive OVX changes if the
one-month lagged excess return of the stock market and that of OVX changes are positive. The sample period is from May 2007 to August 2021.
55
Figure 1. Future value of $1 invested in the trading strategies
This figure plots the future value of $1 invested in the buy & hold, TSMOM, and XTSMOM strategies for all the
countries in May 2007. The strategies include the US one-month risk-free rate. The sample period is from May
2007 to August 2021.
56
Figure 2. Sharpe ratios of the XTSMOM and TSMOM by country
This figure plots the Sharpe ratios of the XTSMOM and the TSMOM for each country. The sample period is from May 2007 to August 2021.
57
Figure 3. Robustness tests for the XTSMOM strategy
0.200 XTSMOM
0.180 TSMOM
0.160
B&H
0.140
0.120
0.100
0.080
0.060
0.040
0.020
0.000
Baseline 12-month lag OVX minus Transaction Local Scaled Orthog.
results stock returns 12-month costs currency XTSMOM XTSMOM
MA
This figure plots the Sharpe ratios for all the countries with the buy & hold, TSMOM, and XTSMOM trading strategies. We show the baseline results as well as the XTSMOM
based on (1) the past 12-month mean stock return as the stock return signal, (2) using the difference between the level of OVX and its 12-month moving average as the oil
uncertainty signal, (3) with the transaction costs of Angelidis and Tessaromatis (2017), 4) the stock market excess returns in local currency, (5) the scaled XTSMOM (see, e.g.,
PSV, 2020), and (6) after orthogonalizing the OVX series. The sample period is from May 2007 to August 2021.
58
Figure 4. The XTSMOM and TSMOM smiles
0.30
XTSMOM
0.25
Strategy excess returns
0.20 TSMOM
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
-0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
Market excess returns
The figure plots the monthly excess returns of the TSMOM and XTSMOM against the corresponding monthly excess
returns of the developed stock market portfolio from Kenneth French’s website. We also plot the second-order
polynomial trendlines for the TSMOM (dashed black line) and the XTSMOM (dotted gray line) monthly returns. The
sample period is from May 2007 to August 2021.
59

Cross-Asset Time-Series Momentum Crude Oil Volatility and Global Stock Markets

Uploaded by

Copyright:

Available Formats

Cross-Asset Time-Series Momentum Crude Oil Volatility and Global Stock Markets

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cross-Asset Time-Series Momentum Crude Oil Volatility and Global Stock Markets

Uploaded by

Copyright:

Available Formats

Cross-Asset Time-Series Momentum:

Crude Oil Volatility and Global Stock Markets

Adrian Fernandez-Pereza, Ivan Indriawanb, Yiuman Tsec, Yahua Xud

We examine the profitability of a cross-asset time-series momentum strategy (XTSMOM)

JEL classifications: G12; G15

returns are negative predictors of bond returns.

In this paper, we incorporate the option-implied oil price volatility in XTSMOM to

companies) postpone irreversible investment decisions in response to increasing oil price

US real GDP growth rate.

markets can be explained by the funding constraints of financial intermediaries. Christoffersen

intermediaries become more capital constrained, leading to lower investment activity.

Therefore, an increase in oil implied volatility indicates a tightening in funding constraints,

strategy. We construct an XTSMOM strategy by examining the predictability of changes in the

superior predictability to a single-asset momentum strategy that relies on only one-month

Christoffersen and Pan (2018).

(increasing) unemployment rates, and lower (higher) inflation.

only partially explained by the various oil shocks.

the predictability of oil implied volatility for global stock returns.

analyses in Section 4. Section 5 concludes.

2. Data and Descriptive Statistics

2.1. Data Sources

our sample period in May 2007 and end in August 2021.

Finally, we collect information on several economic fundamentals. Like PSV (2020),

collected at a monthly frequency from Refinitiv Datastream.

2.2. Descriptive Statistics

changes) and 0.093 (for oil-exporting countries excess returns).

[Insert Table 1 here]

supply/demand shocks. Finally, we report the robustness of our results.

3.1. Single- and Cross-Asset Time-Series Predictability

[Insert Table 2 here]

𝑠𝑖𝑔𝑛(∆𝑂𝑉𝑋) persists. These results are available from the authors.

its two components (𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝐶𝑉)and 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃)). In addition, we control for stock

[Insert Table 2 here]

but 𝑠𝑖𝑔𝑛(∆𝑉𝐼𝑋_𝑉𝑃) is marginally significant at the 10% level in Panels A and C.

negative impact on all our sample countries.

51% of the countries in our sample).

[Insert Table 3 here]

past stock returns combined with positive past OVX changes.

3.2. Cross-Asset Time-Series Strategy

TSMOM strategy excess return, 𝑟𝑡𝑇𝑆𝑀𝑂𝑀,𝑖 , is calculated as follows:

an equal-weighted average of the individual assets’ TSMOM returns—that is, 𝑇𝑆𝑀𝑂𝑀𝑡 =

The XTSMOM strategies are constructed by incorporating the cross-asset predictor as

We form diversified cross-sectional TSMOM portfolios, denoted XTSMOM, by taking equal-

weighted averages of the individual assets’ XTSMOM returns—that is, 𝑋𝑇𝑆𝑀𝑂𝑀𝑡 =

[Insert Table 4 here]

oil-importing and oil-exporting countries in Panels B and C, respectively

( 𝐻4 : 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀(𝑒𝑥𝑝𝑜𝑟𝑡𝑖𝑛𝑔) − 𝑀𝑒𝑎𝑛𝑋𝑇𝑆𝑀𝑂𝑀(𝑖𝑚𝑝𝑜𝑟𝑡𝑖𝑛𝑔) = 0 ). Consistent with our earlier

[Insert Figure 1 here]

[Insert Figure 2 here]

alphas based on the following models:

𝑋𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + 𝛽1 𝑇𝑆𝑀𝑂𝑀𝑡 + 𝛽2 𝑀𝐾𝑇𝑡 + 𝛽3 𝑆𝑀𝐵𝑡 + 𝛽4 𝐻𝑀𝐿𝑡 + 𝛽5 𝑈𝑀𝐷𝑡 + 𝛽6 ∆𝑉𝐼𝑋𝑡 +

𝑋𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + 𝛽1 𝑇𝑆𝑀𝑂𝑀𝑡 + 𝛽2 𝑀𝑆𝐶𝐼_𝑊𝑜𝑟𝑙𝑑𝑡 + 𝛽3 𝑉𝐴𝐿_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 +

𝛽4 𝑀𝑂𝑀_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 + 𝛽5 ∆𝑢𝑠𝑑𝑡 + 𝜖𝑡 , (6)

Datastream.17 𝑉𝐴𝐿_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 and 𝑀𝑂𝑀_𝐸𝑣𝑒𝑟𝑦𝑤ℎ𝑒𝑟𝑒𝑡 are the cross-sectional value and

downloaded from Refinitiv Datastream.18 Finally, 𝜖𝑡 denotes the residuals.

[Insert Table 5 here]

[Insert Table 6 here]

is included in the XTSMOM, but not the reverse.

3.3. Funding constraints of financial intermediaries