Volatility and correlation timing: The role of commodities

• DOI: http://doi.org/10.1002/fut.21939
• Published date: 01 November 2018
• Date: 01 November 2018

Received: 31 October 2017

|

Revised: 14 May 2018

|

Accepted: 14 May 2018

DOI: 10.1002/fut.21939

RESEARCH ARTICLE

Volatility and correlation timing: The role of commodities

Panos K. Pouliasis

|

Nikos C. Papapostolou

Faculty of Finance, Cass Business School,

City, University of London, London, UK

Correspondence

Panos K. Pouliasis, Faculty of Finance,

Cass Business School, City, University of

London, The Costas Grammenos Centre

for Shipping, Trade and Finance, 106

Bunhill Row, London EC1Y 8TZ, UK.

Email: p_pouliasis@city.ac.uk

This paper examines the role of commodities from the perspective of dynamic

asset allocation. We model conditional second moments of stock, bond, and

commodity futures and examine their impact on the portfolio choice decision of

a risk‐averse investor in a mean‐variance framework. Findings suggest that

adding commodities in the opportunity set enhances portfolio risk‐return

characteristics and offers diversification benefits. Moreover, there is substantial

economic value in both volatility and correlation timing strategies. Results are

robust across various subperiods and rebalancing strategies: alternative

correlation dynamics specifications, short‐sale constraints, and transaction

costs under both in‐and out‐of‐sample settings.

KEYWORDS

asset allocation, commodities, volatility timing, correlation timing, multivariate GARCH

JEL CLASSIFICATION

C52, C53, G11, Q02

1

|

INTRODUCTION

Over the last several years, commodity markets have experienced dramatic fluctuations. Significant amounts of funds

allocated to commodity futures and index funds made the sector very popular in the mid‐2000s among institutional

investors of versatile risk attitudes, either as a pure speculation instrument or as a diversification tool. The statistical

features of commodity returns arise from the underlying demand and supply dynamics, yet the price formation function

across commodities is diverse, and this might result in substantial diversification potential.

Investors’interest in commodities is primarily motivated by the belief that commodities offer a hedge against

inflation (Bodie, 1983; Edwards & Park, 1996; Irwin & Landa, 1987) and form an alternative asset class that can

bestow diversification gains to investors. In particular, while equity returns tend to be impacted adversely during

periods of inflation, commodity prices increase and, thus, long positions in commodity futures realize profits. This

is consistent with efficient diversification against downturns in traditional assets such as equity and bond markets

(see Büyüksahin, Haigh, & Robe, 2010; Chong & Miffre, 2010; Gorton & Rouwenhorst, 2006). The diversification

benefits of commodities have been examined by Jensen, Johnson, and Mercer (2000), Belousova and Dorfleitner

(2012), and You and Daigler (2013), among others. For example, Bodie and Rosansky (1980) conduct a

comprehensive analysis of 23 individual commodities during the period from 1950 to 1976 and find that by

switching from a stock‐only portfolio to one that contained 60% stocks and 40% commodities, investors could have

reduced their risk by 30% without giving up any returns. Georgiev (2001) performs a similar study over the period

1995–2005 and demonstrates that adding a commodity component to a diversified portfolio leads to enhanced

Sharpe ratios (SR). Similar are the results of Conover, Jensen, Johnson, and Mercer (2010) who report that

commodity exposure improves portfolio returns in periods of increasing interest rates, consistent with the view

that commodities serve as an inflation hedge.

J Futures Markets. 2018;38:1407–1439. wileyonlinelibrary.com/journal/fut © 2018 Wiley Periodicals, Inc.

|

1407

Another branch of the literature (e.g., Lombardi & Ravazzolo, 2016; Silvennoinen & Thorp, 2013; Tang & Xiong,

2012) argues that the correlation of commodities with stocks and bonds has strengthened. As such, their effectiveness as

an alternative risk‐diversification channel

1

diminishes as a consequence of financialization of the commodity markets.

For example, Daskalaki and Skiadopoulos (2011) challenge their return and risk advantages and find that a mean‐

variance investor is not better off by allocating a portion of their capital to commodities compared to a portfolio that

consists of traditional assets, consistent with the empirical evidence on the increasing financialization of commodities.

Similarly, Cotter, Eyiah‐Donkor, and Potì (2017) implement different strategies and conclude that commodities do not

improve the opportunity set of an investor with an existing portfolio of stocks, bonds, and T‐bills.

Much of the previous research reports mixed evidence on the merits of commodity investment as part of a diversified

portfolio. In essence, these gains are hard to predict and can vary significantly across commodities, throughout time or

with respect to the business cycle. Belousova and Dorfleitner (2012) confirm that there is a strong variation in the

diversification contribution across individual commodities and commodity sectors. This can be attributed to the unique

fundamentals of each commodity sector that makes them uncorrelated with one another. In other words, it is more

meaningful to consider them as a market of separate assets rather than a homogeneous market (e.g., see Erb & Harvey,

2006). In addition, Büyüksahin et al. (2010) find that the alleged benefits that commodities could bring to equity

investors did not materialize when they would have helped the most. This time‐variation in the diversification value is

further confirmed by Adams and Glück (2015) who argue that commodities provide less loss protection after 2008. After

the financial crisis, a new channel transmitting stock market shocks to commodities has opened, especially when the

latter exhibit high volatility. In effect, whether commodities add economic value in asset allocation seems to be linked

to the business cycle and market conditions. For example, Gorton and Rouwenhorst (2006) assert that commodities

improve the risk‐return profile of stock and bond portfolios, and the effect can be more pronounced in late expansion

and early recession phases. Furthermore, Jensen et al. (2000) find that during restrictive phases of the monetary cycle,

commodity futures can lead to significant portfolio return enhancement. Finally, Cheung and Miu (2010) also report

that the diversification gains of commodities are regime‐dependent with the overall long‐run benefits being a result of

the infrequent episodes of outbursts in the commodity markets.

Another reason for conflicting results in the literature might be attributed to the various research designs. The

majority of studies analyzing the contribution of commodity investment in a portfolio of traditional assets is based on

an in‐sample setting. However, in‐sample analyzes implicitly entail forward‐looking information and, therefore, tend to

overstate the achievable gains. For example, Daskalaki and Skiadopoulos (2011) find that commodities contribute only

in‐sample, but do not add value out‐of‐sample. Bessler and Wolff (2015) test different asset allocation strategies and

report that the attainable benefits of commodities are much smaller than suggested by previous studies and depend on

the type of commodity. Other studies conclude that commodities enhance the out‐of‐sample performance of optimized

portfolios (Daskalaki, Skiadopoulos, & Topaloglou, 2017; Gao & Nardari, 2018; You & Daigler, 2013). Given the diverse

conclusions, the out‐of‐sample contribution of commodities remains ambiguous; this constitutes an additional

motivation to explore whether the benefits ascribed to commodities have been exaggerated or not, and investigate the

means to practically exploit them.

The aim of this paper is to empirically examine the impacts of considering commodity investments while at the same

time exploit asset volatility and correlation dynamics from the perspective of dynamic portfolio management. We

consider an active portfolio manager who uses forecasts from dynamic volatility and correlation models to rebalance a

portfolio that contains traditional assets (stocks, bonds, and cash) and a pool of 14 commodities traded on the CME

Group as well as a diversified commodity index. To this end, we compare the performance of different models of

forecasting covariances in terms of optimizing mean‐variance efficient portfolios: (a) sample covariance, (b) constant

conditional correlation (CCC; Bollerslev, 1990), (c) dynamic conditional correlation (DCC; Engle, 2002), (d) mixed data

sampling conditional correlation (MDC; Colacito, Engle, & Ghysels, 2011), and (e) regime switching dynamic

correlation (RSC; Pelletier, 2006). A more accurate set of volatility and/or correlation predictions will render the

investors a way to adaptively adjust their positions so as to achieve a higher utility level. Our analysis aims to provide

market participants with information that can be used to fine tune risk attitudes and support the decision‐making

process.

1408

|

POULIASIS AND PAPAPOSTOLOU

1

Silvennoinen and Thorp (2013) present evidence favoring commodity and financial market integration and document that correlations between stock returns and returns to the majority of

commodity futures have increased. This implies that there might be variables with the capacity to predict both commodity and equity returns (e.g., see Hong & Yogo, 2012). For instance, Asness,

Moskowitz, and Pedersen (2013) find common factors able to explain the pooled cross section of various asset classes including commodities. On the contrary, some earlier studies—prior to the

2007–2009 financial crisis (e.g., Büyüksahin et al., 2010; Chong & Miffre, 2010)—challenge the view of increased integration and argue that commodity returns are affected by commodity‐specific

variables. Hence, equity asset pricing factors cannot explain the cross section of commodity futures suggesting market segmentation (e.g., Bessembinder & Chan, 1992; Erb & Harvey, 2006).

The contributions of this study are several. First, we revisit the role of commodities in asset allocation and their

capacity to provide diversification benefits in a case study that examines portfolio risk‐return characteristics. Results are

validated in terms of SRs and risk‐adjusted abnormal realized returns (Modigliani & Modigliani, 1997). Optimal

portfolios derived from either the traditional asset classes alone (equities, bonds, and cash) or augmented with different

commodity investments. More important, we consider both static and several dynamic asset allocation strategies, and,

therefore, offer additional insights; whether or not the portfolio benefits of commodities depend on the implemented

asset allocation approach. In doing so, we investigate individual commodities and a diversified commodity index

separately, thereby evaluating their potential impact from a portfolio management perspective.

Second, we systematically address the issue under the prism of short‐horizon volatility and correlation timing

strategies. This way, asset allocation efficiency, in terms of risk minimization and return maximization, is directly

linked to predictions of volatilities and correlations. To the best of our knowledge, this is one of a few studies that

explicitly takes into account predictability of second moments in forming optimal portfolios. This aspect has been

largely neglected by asset‐allocation studies that consider commodities that mainly rely on constant historical

estimators (e.g., Belousova & Dorfleitner, 2012; Bodie & Rosansky, 1980; Jensen et al., 2000) or rolling‐sample

estimators (e.g., Bessler & Wolff, 2015; Daskalaki & Skiadopoulos, 2011). An exception is Gao and Nardari (2018) who

consider dynamic forward‐looking strategies. As it is widely agreed that the covariance structure of asset class returns

varies substantially across periods and market conditions, this might have an effect on the diversification value, which

is itself time‐varying.

Third, our analysis focuses not only on whether volatility timing is able to generate economic value compared to a

benchmark strategy but also on any additional value that can be bestowed to the investor when timing both correlations

and volatility. Thus, for the first time to our knowledge, we assess the impact of dynamic correlations separately from

that of volatility and provide a comprehensive analysis of the extent to which dynamic correlations affect optimal

portfolio choice. To capture the trade‐off between risk and return and derive the economic value of dynamic strategies,

we measure the fees mean‐variance risk‐averse investors will be willing to pay to switch from one model to another

based on the postulated utility gains (performance or switching fee); for applications, see Fleming, Kirby, and Ostdiek

(2001; 2003), Della Corte, Sarno, and Tsiakas (2009), and Chou and Liu (2010), among others.

Forth, we assess the robustness of our conclusions to the choice of parameters such as different specifications for

correlation dynamics, rebalancing frequency, estimation period (subperiods), and transaction costs. We also consider

how sensitive our results are to different investment styles—that is, whether there is any impact on the diversification

value of commodities if short selling is not permitted. In addition, since existing studies that support the inclusion of

commodities in the opportunity set are mainly based on in‐sample assessments, we also rely on out‐of‐sample

performance evaluations. Finally, the mean‐variance setting is also contrasted with optimization of alternative risk

measures that focus on tail‐risk (conditional value‐at‐risk [CVaR]).

The structure of the paper is as follows: The next section describes the methodology used to construct optimum

portfolios and quantify volatility and correlation timing gains; Section 3 introduces the econometric methodology and

variance–covariance predictive models; Section 4 presents the data and presents the model estimation results; Section 5

offers the main empirical results on dynamic portfolio management and provides portfolio performance comparisons

based on different models of the conditional second moments; and finally Section 6 concludes the paper.

2

|

OPTIMAL PORTFOLIO SELECTION

In this section, we first formulate the asset allocation problem using mean‐variance analysis. Then, we present the

performance evaluation framework. The details of the methodology are as given below.

2.1

|

Asset allocation in a mean‐variance framework

Our objective is to determine whether there is economic value in conditioning trading strategies on volatility and

correlation, and if so, which specification works the best. For this reason, the standard Markowitz (1952) mean‐variance

portfolio analysis is used. Let +

rt

1

represent the

N

x

1

vector of risky asset returns

,

with conditional expectation

=

++

μ

Er[]

t|t tt

11

and conditional covariance =− −

′

++

+++

H Erμrμ[( )( ) ]

t|t t t t|t ttt

11

111| . For each date

t

, the investor

constructs portfolios through the following optimization:

POULIASIS AND PAPAPOSTOLOU

|

1409