Short‐term salmon price forecasting

• Date: 01 March 2018
• Published date: 01 March 2018
• DOI: http://doi.org/10.1002/for.2482
• Author: Daumantas Bloznelis

RESEARCH ARTICLE

Short‐term salmon price forecasting

Daumantas Bloznelis

1,2

1

Research Centre for Operations Research

and Business Statistics, KU Leuven, Leuven,

Belgium

2

School of Economics and Business,

Norwegian University of Life Sciences, Aas,

Norway

Correspondence

Daumantas Bloznelis, Research Centre for

Operations Research and Business Statistics

(ORSTAT), KU Leuven,Naamsestraat 69,

Box 3555, 3000 Leuven, Belgium.

Email: dbloznelis@gmail.com

Abstract

This study establishes a benchmark for short‐term salmon price forecasting.

The weekly spot price of Norwegian farmed Atlantic salmon is predicted 1–5 weeks

ahead using data from 2007 to 2014. Sixteen alternative forecasting methods

are considered, ranging from classical time series models to customized machine

learning techniques to salmon futures prices. The best predictions are delivered

by k‐nearest neighbors method for 1 week ahead; vector error correction model

estimated using elastic net regularization for 2 and 3 weeks ahead; and futures prices

for 4 and 5 weeks ahead. While the nominal gains in forecast accuracy over a naïve

benchmark are small, the economic value of the forecasts is considerable. Using a

simple trading strategy for timing the sales based on price forecasts could increase

the net profit of a salmon farmer by around 7%.

KEYWORDS

elastic net regularization, financial timeseries, forecasting, k‐nearestneighbors, salmon price, vector error

correction model

1|INTRODUCTION

Forecasting the future spot price of salmon is a bread‐

and‐butter activity of the salmon market participants.

Salmon farmers tailor their output schedule to match the

profit‐maximizing production path, which depends on the

expected trajectory of the spot price. Salmon processors plan

their operations relying on expectations of their input cost

level, which is determined by the future spot price of salmon.

Traders and intermediaries between producers, processors

and retailers are especially sensitive to the current and future

spot prices through the cost of buying the commodity and the

revenue from selling it. Hedgers and speculators in the

salmon futures market make or lose money depending on

how good are their forecasts of future spot price. Hence the

importance of price prediction is hard to overstate.

Nevertheless, previous studies of salmon price forecast-

ing are scarce. Lin, Herrmann, Lin, and Mittelhammer

(1989) considered Norwegian salmon prices in the USA

and the European Community (EC) using monthly data for

1983–1987. A structural dynamic simultaneous equation

model together with Box–Jenkins time series techniques

was used to predict the monthly price of salmon in the USA

and the EC for 1989–1992. The forecasts were not evaluated

as they targeted future prices that were unavailable at the time

of publishing the article. Vukina and Anderson (1994) fore-

casted wholesale prices of five wild salmon species in Japan

employing state‐space modeling techniques on monthly data

for 1978–1991. Although they obtained gains over naïve no‐

change forecasts, their forecast accuracy varied widely across

species and forecast horizons. Also, their hold‐out sample

consisted only of six time points, which is arguably too few

for making reliable inference on forecast accuracy. Gu and

Anderson (1995) used monthly data for 1986–1993 in a

state‐space model similar to that of Vukina and Anderson,

to forecast both wild and farmed salmon prices in the US

wholesale market. Predictions for 3–12 months ahead were

satisfactory in terms of mean absolute percentage error and

correctly indicated the direction of the price change around

60–67% of the time. Finally, Guttormsen (1999) forecasted

weekly prices of farmed Atlantic salmon organized into

weight classes using data for 1992–1997. Autoregressive

integrated moving average (ARIMA), vector autoregression

(VAR), and Holt–Winters exponential smoothing methods

Received: 3 November 2016 Revised: 6 March 2017 Accepted: 26 May 2017

DOI: 10.1002/for.2482

Journal of Forecasting. 2018;37:151–169. Copyright © 2017 John Wiley & Sons, Ltd.wileyonlinelibrary.com/journal/for 151

were used, among others. Four‐,8‐, and 12‐week forecasts

were all rather accurate in terms of mean absolute percentage

error and proportion of accurate forecasts of the direction of

price change. However, even the best model did not improve

upon a naïve forecast when the most representative weight

classes were considered.

All the above studies on salmon price forecasting are

obsolete from an empirical point of view; hence I find the

time ripe for an update. Currently, a new take on forecasting

the price is especially relevant as the price volatility measured

by the standard deviation of logarithmic returns on the

weekly spot price doubled in 2005–2007 (Bloznelis, 2016b;

Oglend, 2013). The volatility has stayed persistently high

ever since, bringing increased uncertainty about future pr ice

development and unsettling salmon market participants

(Jensen, 2013).

I will explore a number of forecasting techniques in

pursuit of the best salmon price prediction 1–5 weeks ahead.

I will assess the forecasts using absolute and relative accuracy

measures. Implications of findings with regard to salmon

market participants will be discussed. The study will serve

as a benchmark for salmon price forecasting and will

highlight forecasting methods that deserve increased

attention in future works.

The remainder of the paper is organized as follows.

Section 2 discusses the salmon market and factors determin-

ing the spot price of salmon. Section 3 motivates the selec-

tion of variables and presents the data. Section 4 provides

descriptions of forecasting techniques and forecast evaluation

methods. Section 5 explores the forecasting results and

compares them across techniques. A discussion on the

implications of findings concludes the paper.

2|THESALMONMARKET

Salmon farming is a major industry in Norway, with output

slightly above 1 million tonnes valued at around NOK 45

billion in 2014 (The Fish Site, 2015).

1

Production is pre-

dicted to expand by 3% annually over 2013–2020 (Marine

Harvest, 2014b). Norway is the largest farmed salmon pro-

ducer in the world, with over 60% of the global production

volume as of 2014 (Marine Harvest, 2014a); it exports nearly

all of its production and is the major salmon supplier for

Europe, the largest salmon consumer continent. Other large

salmon farming countries are Chile, Canada and the UK.

Norwegian salmon is traded in the spot market and using for-

ward contracts.

Salmon spot price results from the interplay of supply and

demand. The supply volume is limited by the standing

biomass and slaughterhouse capacity from above, and natu-

ral, biological, and regulatory factors from below. Biomass

is the total amount of fish in the farming facilities. It evolves

relatively slowly due to continued slaughtering throughout

the year and natural limitations on the speed of growth of

salmon in the seawater. The speed of growth depends on

seawater temperature and feeding intensity. Biomass is highly

seasonal and peaks around September each year (Mar ine

Harvest, 2014a). Industry regulations limit the maximum

biomass at each farming site; once the limit is reached, the

farmer must start slaughtering or otherwise pay a fine.

Among the biological factors, infectious diseases and

parasites play an important role. When the fish cannot be

cured or this is too expensive, they are slaughtered and sold

(most salmon diseases are not dangerous to humans). For

example, an outbreak of infectious salmon anemia (ISA) in

Chile decimated the salmon biomass in 2007–2010. This

caused considerable fluctuations in supply and changed the

global trade flows of salmon; see, for example, Asche,

Hansen, Tveterås, and Tveterås (2009).

Sexual maturation is another determinant of the salmon

supply. It takes place once the fish reach a certain age and

results in inferior flesh quality. Hence farmers aim to

slaughter the fish before it occurs. Sexual maturation is a

highly seasonal phenomenon, normally occurring in late

summer and autumn, but it may be accelerated or delayed

depending on feeding and other factors (Pall, Norberg,

Anderson, & Taranger, 2006).

Salmon demand fluctuations are mainly seasonal.

Christmas and Easter are prime examples of increased

demand periods. In the long run, consumer tastes and

purchasing power are the main drivers of demand.

In addition to the fundamental factors, cur rency exchange

rate is an important element in the supply–demand relation-

ship in the spot market for salmon. Almost all Norwegian

farmed salmon is exported (Marine Harvest, 2014a); while

consumers face retail prices in foreign currencies, producers'

home currency is NOK. Bloznelis (2016b) found the EUR/

NOK exchange rate to be an economically significant

determinant of the spot price in 2007–2013 (see also

Straume, 2014).

With the launch of the Fish Pool futures and options

exchange for salmon in 2006, a new medium of price discov-

ery appeared. Daily price quotes on monthly contracts from 1

to 60 months ahead became publicly available. The trade on

Fish Pool generates new information on the future develop-

ment of the salmon spot market. Price quotes for each con-

tract have to be published daily regardless of whether there

are any trades on the contract. On days with no trades (which

are not that rare) the analysts at Fish Pool use their best judg-

ment to reflect any relevant information in making up the

quotes. This helps in timely dissemination of information

and may influence price formation in the spot market.

1

To obtain production volume from export volume, I add 3–4%, which

accounts for local consumption (Marine Harvest, 2014a).

152 BLOZNELIS