A tale of two premiums revisited

• Published date: 01 May 2023
• Author: Loïc Maréchal
• Date: 01 May 2023
• DOI: http://doi.org/10.1002/fut.22396

Received: 31 December 2022

|

Accepted: 2 January 2023

DOI: 10.1002/fut.22396

RESEARCH ARTICLE

A tale of two premiums revisited

Loïc Maréchal

Department of Information Systems,

HEC Lausanne, University of Lausanne,

Cyber‐Defence Campus, armasuisse

Science and Technology, University of

Lausanne and Cyber‐Defence Campus,

armasuisse Science and Technology,

Lausanne, Vaud, Switzerland

Correspondence

Loïc Maréchal, University of Lausanne,

Internef, Quarter UNIL‐Chamberonne,

Lausanne, Vaud, CH‐1015, Switzerland.

Email: loic.marechal@unil.ch

Abstract

This paper investigates the effect of the “financialization”of commodity markets

in terms of pricing. I explore whether the emergence of commodity index

traders (CITs) affects weekly returns and turnover during the roll periods. I split

the sample (1994–2017) into prefinancialization (1994–2003) and postfinancializa-

tion (2004–2017). I directly test whether the CIT market share (CIT/open interest)

contributes to commodity returns and whether risk adjustments (based on

momentum, basis, basis‐momentum, open interest, crowding, and average factors)

alter liquidity and insurance premiums documented in Kang, Rouwenhorst, and

Tang. I also examine how the financialization affects liquidity and insurance

premiums. Finally, since previous results are obtained with Fama–MacBeth

regressions, I use an alternative method to test how liquidity and insurance

premiums determine commodity returns.

KEYWORDS

commodity futures, financialization, hedging pressure, liquidity premium, risk premium

JEL CLASSIFICATION

G13, G14, G23

1|INTRODUCTION

In a recent paper, Kang et al. (2020) (hereafter, KRT)uncover two risk premiums in commodity returns, that is, the long‐

term horizon insurance premium paid by producers through hedging demand, and the short‐term horizon liquidity

premium earned by these producers for easing the trading activity of speculators. The goal of this paper is to check the

robustness of these results under different scenarios. I conjecture that a third category of investors, the commodity index

traders (CITs), affects these premiums since their objective is to get exposure to the spot price of commodities rather than

just grabbing the insurance premium. These investors are net “buyers”of commodity contracts. On the one hand, this

should decrease the insurance premium, and on the other hand, CITs rolling their position from the nearby to higher

maturity contracts should increase the liquidity premium on the nearby contract, at least during the period starting after

the roll to the maturity of the contracts. The net effect of CITs on contract demand mirrors that of processors and users.

However, the causes of this demand are different. While the former include commodities in their portfolio for

diversification purposes (wealth management), the latter are hedging their position (risk management).

Testing for the existence of (risk) premiums in commodity futures markets is a long‐lasting challenge. The reasons

include, among others, the small number of contracts available (cross‐section), their heterogeneity in terms of demand,

J Futures Markets. 2023;43:580–614.580

|

wileyonlinelibrary.com/journal/fut

This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any

medium, provided the original work is properly cited, the use is non‐commercial and no modifications or adaptations are made.

© 2023 The Authors. The Journal of Futures Markets published by Wiley Periodicals LLC.

their lack of integration with traditional asset classes, and nonavailable spot prices. From a theoretical perspective,

Black (1976) denies the existence of a commodity risk premium resulting from exposure to a global factor. A large

strand of the empirical literature supports this view when additional commodity market peculiarities (e.g., hedging

pressure) are not controlled appropriately; see, for example, Dusak (1973), Bodie and Rosansky (1980), or Carter et al.

(1983). Moreover, whereas it seems natural to relate commodities returns to changes in aggregate consumption, and

despite the existence of theoretical models (consumption‐based capital asset‐pricing model [CAPM]) designed for

commodity pricing, there is weak empirical evidence concerning the existence of a significant risk premium; see, for

example, Hazuka (1984) and Jagannathan (1985). Empirical tests built on characteristics derived from the theory of

“Normal Backwardation”or the theory of “Storage”do not provide more convincing results; see Keynes (1930), Hicks

(1946), Kaldor (1939), Working (1949), and Brennan (1958). Similarly, characteristics derived from market frictions

common to many asset markets, such as transaction costs, liquidity, or limits to arbitrage, also work well to explain the

cross‐section of commodity futures returns. This research uses these characteristics and traded factors built upon these

characteristics to provide risk adjustments.

To identify under which conditions the results of KRT hold after the inclusion of variables related directly or

indirectly to the financialization, I determine the set of factors that allows for an optimal risk adjustment. Beyond

insurance and risk premiums, I use six additional characteristics known to predict commodity futures returns. Next,

from these characteristics, I construct factors (long–short portfolios sorted on characteristics and rebalanced weekly).

As previously documented for monthly or bimonthly frequencies, I document a superior performance for portfolios

sorted on (i) basis (annualized average weekly returns of 15.75%), (ii) momentum (14.46%), and (iii) basis‐momentum

(14.04%). I also confirm their performance at the weekly frequency for portfolios sorted on (i) average hedging pressure

(9.99%), (ii) net trading of “commercials”(17.99%), and (iii) crowding (17.85%); see Kang et al. (2020,2021). In contrast,

the portfolios sorted on the

β

with an average portfolio (Bakshi et al., 2019) or on the growth of open interest (H. Hong

& Yogo, 2012) deliver poor annualized average weekly returns (

−

0.5

8

% and

−

1.5

1

%, respectively).

Next, I select the optimal subset of factors with the Bayesian asset‐pricingtest derived by Barillas and Shanken (2018).

The initial set of factors does not include the characteristics under scrutiny (average hedging pressure and net trading).

The highest probability (61%, for a prior on the maximum attainable Sharpe ratio of 1.25) is obtained with the four‐factor

model that includes the basis, momentum, basis‐momentum, and crowding factor. This result holds when the test assets

are the remaining portfolios or the remaining portfolios plus the 26 individual commodities of the sample.

First‐generation indices have a systematic methodology to carry their long position from the nearby onto the first

deferred futures contract called the “roll.”The amount rolled every day over the window is known.

1

Using panel data

and time series regression, I perform preliminary analyses concerning the impact of the roll‐days on commodities

returns and turnover panels and on the factor returns, respectively. I do not find any statistically significant systematic

effect of the roll‐days on individual commodity returns, and on the factors, in both pre‐and postfinancialization

periods. This result supports the view that the financialization does not affect prices by rolling SP‐GSCI contracts.

Conversely, I find that the roll is associated with a significant increase in turnover, in particular for the nearby contract.

Moreover, this effect is the most visible in the postfinancialization period, with an average cumulative abnormal

turnover of 17.56%.

Next, I replicate the analysis of KRT (Fama–MacBeth regressions, Fama & MacBeth, 1973). I also estimate a related

model where I introduce the optimal set of factors selected previously. I find that the insurance (average hedging

pressure) and liquidity (net trading) premiums are robust to this adjustment, despite a drop in the economic magnitude

(from 0.43 to 0.34) and the statistical significance (from 1% to 10%) of the insurance premium. A one standard deviation

change in AH

P

impacts the returns by 8.2 and 6.5 bps, respectively.

I pursue these robustness tests and control for the financialization with three different approaches. First, I add

a direct measure of index funds pressure to the regression. I find that KRT results are not affected by this variable

and that the coefficient of this variable does not show any statistical significance at the usual levels. Second,

I restrict the Fama–MacBeth regressions to the cross‐sections of the weeks with at least a 3‐day overlap with the

roll of the SP‐GSCI. This aims to control whether KRT results are not driven by the particular roll weeks. The

significance levels of the insurance and liquidity coefficients drop at the 5% and 10% levels, respectively. In

addition, while the economic magnitude of the insurance coefficient is unchanged, that of

Q

is halved at 2.32%.

Third, I split the sample into a pre‐and postfinancialization period to capture more generally a change in the

1

From the fifth to the ninth workable days for the SP‐GSCI.

MARÉCHAL

|

581

risk‐sharing structure based on the study period. I find that the liquidity coefficient is not affected by the period,

butthatoftheinsuranceisnotsignificantattheusuallevel(significantatthe10%level)inthepre‐(post)

financialization period.

Several papers have documented the econometric limits of the Fama–MacBeth approach; see Petersen (2009) and

Gow et al. (2010). Therefore, I use an alternative approach developed by Hoechle et al. (2020).

2

It permits the inclusion

of contract fixed effects (FEs), a feature that seems necessary, given the important heterogeneity of commodities. It

additionally allows controlling for the dependence between the construction of risk‐adjustment factors and the

characteristics of interest (or unobservable characteristics). In this setting, I confirm that contract FEs contribute

significantly to explaining the returns. The Haussman (1978) test rejects the random effect hypothesis at the 1% level in

all specifications. However, I find that the liquidity premium is robust to econometric and risk adjustments but not the

insurance premium. When FEs are added, the coefficient is not significant at the usual levels. Moreover, this latter

coefficient becomes negative (and not statistically significant) in the postfinancialization period. Finally, I conclude the

analysis with an economic robustness specification. The disaggregated CFTC data set, available since 2007, allows the

computation of more accurate values of CITs pressure. I do not find any statistically significant effect related to CITs

pressure, and confirm that in the postfinancialization period, additionally extended until 2020, the insurance price is

not statistically significant either.

My contribution is threefold. First, I identify factors that optimally price commodity futures returns. This has

implications in research to, for example, improve the counterfactual of event studies or benchmark the

performance of future factors. Contrary to Boons and Porras Prado (2019), who compare the performance

of the basis‐momentum in turn, against the basis and Bakshi et al. (2019) factors (momentum, basis, and average

factor), I find that, at the weekly frequency, basis‐momentum does not suffice to generate the optimal asset‐

pricing model. This result is also useful for portfolio management in practice. Second, I show that the roll of index

funds neither affects the returns nor the factors. However, I find that the roll‐days increase the turnover, implying

that CITs modify the functioning of commodity markets. Finally, using an appropriate methodology for

longitudinal panels, I find that the financialization does not affect the liquidity premium earned by producers,

whereas the price of the insurance premium decreases and eventually vanishes. This goes against the argument

that index funds activity is detrimental to the commodity economic activity since it benefits commodity

producers.

The remainder of the paper is organized as follows. Section 2provides a literature review and develops hypotheses

on previously identified characteristics, risk premiums, and integration of commodity markets in the broader economy.

Section 3describes the data and presents the summary statistics. Section 4presents the empirical results, and Section 5

provides the econometric robustness tests. Section 6concludes.

2|LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT

2.1 |Theoretical models of commodity prices

2.1.1 |Normal backwardation

The theory of Normal Backwardation states that producers depress the futures price below the expected spot price

because they sell future output through futures contracts to hedge against price fluctuations; see Keynes (1930) and

Hicks (1946). Hence, a long futures investor obtains a positive premium from the producer that insures herself. Cootner

(1960) and Gray (1961) extend the theory to buyers of physical commodities who generate net‐long hedging pressure

and potentially a sign reversal of the insurance premium. Testing this theory is a threefold question: (i) Is the futures

price a downward biased estimate of the expected spot price? (ii) Are speculators' profits positive on average? (iii) Do

speculators' profits arise from the insurance premium or superior forecasting ability? A subsequent strand of literature

builds on the real or latent substitution effects across commodities, leading to cross‐hedging. Many physical

commodities do not have corresponding futures markets. Therefore, commodity hedgers may use a resembling or a

linear combination of resembling futures contracts to hedge their position; see Anderson and Danthine (1981). If the

2

An obvious limitation is the limited cross‐sectional dimension, of 26 commodity contracts, in this context.

582

|

MARÉCHAL