The predictability of aggregate returns on commodity futures

• Published date: 01 September 2014
• Date: 01 September 2014
• Author: Fabian T. Lutzenberger
• DOI: http://doi.org/10.1016/j.rfe.2014.02.001

The predictability of aggregate returns on commodity futures

Fabian T. Lutzenberger ⁎

FIM ResearchCenter Finance & InformationManagement, Universityof Augsburg, 86135 Augsburg,Germany

Instituteof Materials ResourceManagement, Universityof Augsburg, 86135 Augsburg,Germany

abstractarticle info

Articlehistory:

Received9 December 2013

Receivedin revised form 17 January 2014

Accepted3 February 2014

Availableonline 8 February 2014

JEL classification:

G12

G13

G17

Keywords:

Asset pricing

Commodities

Predictabilityof returns

Predictiveregressions

Forecasting

This paper providesevidence that aggregate returns on commodityfutures (without the returns on collateral)

are predictable, both in-sample and out-of-sample, by various lagged variables from the stock market, bond

market, macroeconomics, and the commoditymarket. Out of the 32 candidate predictorswe consider, we find

that investorsentiment is the best in-sample predictorof short-horizon returns,whereas the level and slope of

the yield curve have much in-sample predictive power for long-horizonreturns. We find that it is possible to

forecast aggregate returns on commodity futures out-of-sample through several combination forecas ts (the

out-of-samplereturn forecasting R

2

is up to 1.65%at the monthly frequency).

© 2014 ElsevierInc. All rights reserved.

1. Introduction

Are asset returns predictable? Th e 2013 Nobel Prize recipients

Eugene Fama,Lars Peter Hansen and Robert Shiller find that this ques-

tion “is as central as it is old”(The Royal Swedish Ac ademy of

Sciences,2013). While several studiesexamine whether excessreturns

on assetclasses such as stocks,treasuries, bonds, foreignexchange, sov-

ereigndebt, and housesare predictable (Cochrane,2011 and the articles

cited therein),there is less (up-to-date) evidencefor the predictability

of commodityfutures returns, despitethe fact that “commodityfutures

have [by now]moved into the investmentmainstream”(Basu& Miffre,

2013).

This paperattempts to fill this researchgap to some extent by study-

ing thepredictability ofaggregate returnson commodity futures,that is,

we test whether the null of unpredictable commodity futures returns

can be rejected and seek to identify va riables that show predictive

power. For thispurpose, we do not empiricallytest one specific theory

of commodityfutures returns andits implications, as the theoryof stor-

age of Kaldor (1939) and others.Instead, we follow an empirical asset

pricingapproach. Thatis, by working backwards,we come from the em-

piricalside and test a large setof potential predictors.Most of these can-

didatepredictors are standardchoices in studies ofreturn predictability

of otherasset classes, especiallythat of stocksand bonds. In addition,we

propose somenew factors, of which most are commodity-specific. The

empiricalfacts we identifyshould be subject to furthertheoretical stud-

ies thatseek to propose theoreticalmodels that capturethese given em-

pirical patterns.

1

To be somewhat more specific, we conduct predictive regressions

that use the future return on an equal-weighted portfolio of 27 com-

modity futures(without the return on collateral) as the response vari-

able. The right-hand sides of these re gressions comprise the current

values of subsets of 32 potential pred ictors from the stock market,

bond market, macroeconomics,and the commodity market. The main

sample periodis from January 1972 to June2010, with a monthly sam-

ple frequency. Predictiveregressions arethe most common approachto

forecast aggregate returns (Kelly& Pruitt, 2013). If returns are unpre-

dictable, regressioncoefficients beyond a constant should be insignifi-

cant in such models,and these models should not provide forecasts of

future returns that are more ac curate than the historical av erage of

past returns. We evaluate both th e in-sample (IS) and the out-of-

sample (OOS) predictability. For the IS an alysis, we employ single

long-horizon predictive regressions with hor izons of 1, 3, 12, 24, 36,

48, and 60 months ahead (Maio & Santa-Clara, 2012; amo ng others)

as well as a procedure that selectsthe “best”multiple-variable regres-

sion out of the variety of candidate predictors we consider, as is pro-

posed by Bossaerts and Hillion (199 9) and Zakamulin (2013).The

Reviewof Financial Economics 23 (2014)120–130

⁎Tel.:+ 49 821 5984884, + 49821 598 4801 (Secretariat);fax: +49 821 598 4899.

E-mailaddress: fabian.l utzenberger@wi wi.uni-augsb urg.de.

1

Fama and French(2013) describe the researchphilosophy of empiricalasset pricing.

1058-3300/$–see front matter © 2014 ElsevierInc. All rights reserved.

http://dx.doi.org/10.1016/j.rfe.2014.02.001

Contents listsavailable at ScienceDirect

Review of Financial Economics

journal homepage: www.elsevier.com/locate/rfe