Old crop versus new crop prices: Explaining the correlation

• Published date: 01 July 2020
• DOI: http://doi.org/10.1002/fut.22106
• Author: Francisco Arroyo Marioli
• Date: 01 July 2020

J Futures Markets. 2020;40:1192–1208.wileyonlinelibrary.com/journal/fut1192

|

© 2020 Wiley Periodicals, Inc.

Received: 16 July 2019

|

Accepted: 28 January 2020

DOI: 10.1002/fut.22106

RESEARCH ARTICLE

Old crop versus new crop prices: Explaining the correlation

Francisco Arroyo Marioli

Central Bank of Chile, Metropolitan

Region, Santiago, Chile

Correspondence

Francisco Arroyo Marioli, Central Bank of

Chile, Metropolitan Region, Agustinas

1180, Santiago, Chile.

Email: farroyomarioli@g.ucla.edu and

farroyo@bcentral.cl

Abstract

Although competitive storage theory has proven successful in explaining many

patterns for commodity prices, some features are not understood. While

standard models predict low correlation between future prices with delivery

dates before and after the harvest, the data suggest otherwise. To correct this, I

assume that harvests appear continuously rather than at a single moment. This

addition to the standard model allows me to link preharvest and postharvest

markets to the same source of supply, and hence obtain the empirically ob-

served high correlation. Empirical evidence also suggests that my assumptions

are realistic. Results are robust to different parameter specifications.

KEYWORDS

backwardation, commodities, correlation, futures, seasonality

1|INTRODUCTION

Commodity markets have several characteristics that define them and render them unique. They are more homo-

geneous, more transparent, and more liquid than other types of goods. Information regarding price and quantity is

usually available at high frequencies. Many of them have markets with delivery dates in the future, as well as call and

put options. Compared with markets, like, manufacturing, the market for commodities is much more developed.

Because they are mostly natural resources, the production process also tends to be unique. The extension of this paper

will be focusing on farming products to explore the implications of a key feature of agricultural markets: the seasonality of

their production process. By seasonality, I will include any production process that presents exogenous monthly variations

throughout the year and systematically every year. That is, seasonal production processes are variations that repeat

themselves every year and hence can be forecastable within a certain range. Differences in seasonality allow for different

future curves, depending on the product in question, its harvest season, and the geographic location of markets.

Agricultural goods differ from most other goods because they can only be produced (harvested) during a certain interval

of the year. Production decisions must follow a certain timing and schedule. This process is common knowledge and allows

for a set of future contracts to be signed before delivery dates. Since production is irregular, but demand for these goods is

steady throughout the year, the inevitable answer is to add storage to the industry. Within a closed‐economy framework, the

existence of storage means that throughout the year consumers will be “eating”old‐harvest leftovers (stored in silos) while

waiting for the “new”harvest to come in at some point. Therefore, future prices with delivery dates before and after the new

harvest will have different sources of supply. Since supply at different moments in time will come from different sources

(harvests), supply shocks (or information shocks regarding supply) should not “move”future prices in the same direction.

Moreover, shocks regarding the new harvest (a USDA report that forecasts the next harvest) should only affect futures with

postharvest delivery dates whenever these are in backwardation, which is the usual case for agricultural commodities since it

is common to see a drop in future prices that mature after the next harvest. However, the data show the opposite.

Correlations between new and old future prices for all main agricultural commodities are not only positive but also

consistently high throughout the years, usually between 0.7 and 1. This can be better seen in Supporting Information

Table A.1, which shows the correlation between first generic futures (futures that mature closest to date) and fifth generic

ones (fifth future contract ordered by maturity) for four crops: corn, cotton, wheat, and soybeans. The first explanation for

this is that demand shocks might be autocorrelated, and hence allow for positive correlation between all future prices.

However, Pirrong (2011) deems this hypothesis incomplete by analyzing the correlations under supply‐side news shocks.

That is, he studies correlation only in the case of shocks regarding supply (e.g., a change in the harvest forecast), and finds

that even after conditioning the sample to supply‐related information shocks, correlation is still positive and close to one.

Therefore, some supply based explanation is still required.

The main hypothesis in this paper is that new harvests are sold in both “old”and “new”markets. That is, if we start

by assuming that harvests come “in pieces”(instead of all in one moment, as the literature does), a market equilibrium

will result in selling the “early part”of the harvest in the old markets and the rest in the new one. Hence, both markets

would have a common source of supply, allowing for supply induced correlation.

It is also important to state that the model used in this paper is consistent with the standard literature. Competitive

storage models have been widely used in the commodities literature, with many positive results when contrasted

empirically. The main goal here is to explain correlations between future prices of seasonal goods without contradicting

the key findings competitive storage models have already obtained.

The paper is organized as follows. Section 2details the main findings of the literature and how this issue has been

approached thus far. Section 3introduces the model, explains the puzzle I address, and presents theoretical results.

Section 4presents the empirical evidence for the assumptions used and results obtained, and Section 5concludes. The

Supporting Information includes proofs for the results from Section 3.

2|LITERATURE OVERVIEW

Storage theory has been widely used throughout the commodities literature. It dates to almost a century ago, with

Kaldor's (1939) convenience yield hypothesis that future prices are expected spot prices adjusted for storage costs,

opportunity costs, and an implicit benefit: convenience yield. The presence of convenience yields to explain future

curves has been widely used in the literature (Brennan, 1958; Pindyck, 1994; Pindyck, 2001; Telser, 1958; Turnovsky,

1983; Working, 1949). This theory contrasts with Keynes's (1930) normal backwardation theory, which describes future

prices as expected spot prices plus a risk premium. Other papers explore alternatives with other basis than back-

wardation nor risk premia, such as Brennan, Williams, and Wright (1977), Samuelson (1957,1971) and Williams and

Wright (1989). However, empirical analysis of competitive storage theory did not begin until the late 1980s, probably

due to the availability of data and computational capacity. Many statistical aspects of these markets have been studied

within this framework. Fama and French (1987) test storage models and analyze the relation between the basis (the

difference between future and spot prices) and risk premiums (a normal backwardation approach). They find evidence

to support storage theory and also, for some commodities, evidence in favor of the risk premium approach. In the same

direction, Gorton, Hayashi, and Rouwenhorst (2013) also find evidence for the predicted relation, both between

inventories and basis and between inventories and risk premiums, in a much broader set of commodity data. They do

not, however, find a relation between trading position and risk premiums. Both Ng and Pirrong (1994) and Fama and

French (1989) analyze variability in spot and future prices and obtain their relation with inventory levels, by comparing

model predictions with data, and again obtain interesting results in favor. Many other papers resulted from tested

empirically previously developed theory (Carter & Revoredo Giha, 2007; Schwartz, 1997).

In a seminal paper, Deaton and Laroque (1992) apply a standard rational expectation competitive storage model to

13 commodities and match moments in the data, such as skewness, kurtosis, and conditional variances depending on

the type of demand shocks simulated. Dvir and Kenneth (2014) apply a storage model with growth to the oil market and

obtain results that match the data in certain respects for both before and after 1973, when supply became restricted.

More related to the issue addressed in this paper, Pirrong (2011) examines correlations between new‐and old‐harvest

prices for six agricultural commodities for a 30‐year sample and shows a systematically high and positive correlation for all

goods. Moreover, he repeats this analysis but then conditions the sample on USDA harvest forecast report release dates; that

is, he only tests for correlations on days when such reports are released. These USDA harvest reports contain information

regarding supply: harvest forecasts, quality conditions, expected timing, and so forth. These results conditioned on report

release dates still show a high positive correlation. Therefore, one cannot simply explain such correlation through auto-

correlated demand shifts, as in Deaton and Laroque (1992). They demonstrate that these results are inconsistent with

simulated results from a storage model, and also claims that alternative explanations, such as intertemporal substitution and

ARROYO MARIOLI

|

1193