Estimating prices for “new” aquaculture species: A hedonic pricing approach

Abstract

Assessing the potential economic and financial feasibility of a new aquaculture operation requires, amongst other things, an estimate of the price the product is expected to receive once produced. For new species, such prices may not be apparent in the market. In this study, we develop a hedonic pricing model to estimate the expected price of a species being considered for aquaculture, but which does not have a current market presence in Australia. The model includes information on a wide range of other species that are currently traded in the market, including low and high value species; farmed and wild caught, and domestically produced or imported. The characteristics of these species, including product form (e.g., whole, fillets, fresh, frozen, etc.) and taste characteristics (e.g., texture, moistness) are regressed against their price. Four different functional forms of the model are applied: log-linear, linear, semi-loglinear and the use of a Box-Cox transformation of the price variable. Sensory testing was undertaken to assess the characteristics of the new species, and the model was used to provide estimates of its likely market price if it is to be produced.

Keywords:

• Australian aquaculture
• fish characteristics
• hedonic price model
• sensory testing

Introduction

The growth in aquaculture since the 1960s had largely been driven by the introduction of new species and increases in productivity that have seen production costs of these species decline over time (Asche et al., 2022). In most cases, such as for key aquaculture species such as shrimp, salmon, tilapia, barramundi and catfish, the species have had a preexisting presence on the market, and the challenge for aquaculture was to develop methods to produce these species in a controlled production process in a cost-effective manner.

In Australia, interest has recently developed in the potential for aquacultural production of Trachinotus anak, a tropical fish species caught in northern Australia waters. The species, which has the official standard fish name of Giant Oystercracker Dart in Australia (FRDC, No date), is caught in small quantities by recreational fishers, with some catch taken by commercial fishers as incidental byproduct (between 1 and 5 tonnes a year). The species is related to the Florida pompano (Trachinotus carolinus) that is currently being farmed in the USA (Weirich et al., 2021), and the Snubnose dart (Trachinotus blochii) and golden pompano (Trachinotus ovatus) farmed in south-east Asia (McMaster & Gopakumar, 2022; Shen et al., 2021; Xiang, 2015), suggesting the potential viability of similar production in Australia.

A critical missing component to assess the economic feasibility of farming the species is the likely price the product will obtain once it reaches the market. Although some commercial catch of T. anak and the related species, T. blochii, is sent to the market, this occurs irregularly and in small quantities, and hence the price received is unlikely to represent the price that would be received for larger quantities of aquaculture production.

In this study, we develop a hedonic pricing model for key fish species sold on the Australian market. The study differs from most previous seafood related applications in that the model not only estimates the price of a species based on its set of readily observable characteristics, such as origin, sustainability certification or product form, but also on its less observable but intrinsic taste and texture characteristics. Sensory testing was undertaken to determine the latter set of characteristics of T. anak, as well as other species for cross comparison. From this, we estimate the likely farm-gate price of the potential aquaculture species at different levels of production.

Materials and methods

Hedonic price modeling

Hedonic pricing is predicated on the theory that the price of a good reflects the set of its intrinsic characteristics (Lancaster, 1966; 1971), with econometric models to estimate the contribution of these characteristics to price developed first by Rosen (1974). Since then, applications of hedonic pricing models have predominantly been applied to housing data, often with the aim of identifying the contribution of environmental attributes to house prices (e.g. Garrod & Willis, 1992; Nicholls, 2019) or the value of access to services such as education or transportation (e.g., Hawkins & Habib, 2018; Luo et al., 2021). Evans et al. (2017) used the approach to determine the impact of marine aquaculture development on coastal housing prices.

Hedonic pricing models have also been applied to food markets (Costanigro et al., 2011), including those of fish species, focusing on particular attributes. For example, Asche et al. (2021) used hedonic pricing to estimate the value of sustainability certification to farmed whitefish, while other studies (e.g. Hammarlund, 2015; Hukom et al., 2020; Kehoe et al., 2023; Kristofersson & Rickertsen, 2007; Lee, 2014; McConnell & Strand, 2000; Pettersen & Asche, 2020; Ray et al., 2022; Roheim et al., 2007) used the approach to determine the contribution of attributes such as fish size, freshness and product form to price. Other studies have considered the impact of the market structure itself on prices received. For example, Wolff and Asche (2022) estimated the effects of transaction mode (i.e., auction or direct sales) on price while Ray et al. (2022) identified separate submarkets for salmon species based on species and production method (i.e., farmed versus wild caught).

Observed market prices are described by the intersection of both supply and demand, and that the set of characteristics determining the price are not necessarily sufficient to identify either demand nor supply (Rosen, 1974). Despite this, a single equation approach is most commonly used, with price expressed as a function of the characteristics of the good(s) considered. While most studies utilize a cross section of data, Arguea and Hsiao (1993) suggest that temporal cross-sectional data (i.e., panel data) are required to improve the identification of the demand component, such as used by Wolff and Asche (2022) and Ray et al. (2022).

Underlying the use of a single equation approach is the assumption that supply is highly inelastic, in that price changes reflect shifts in demand only, adjusting to clear the market (McConnell & Strand, 2000). In such cases, the quantity supplied can be considered exogenous, and a single equation approach is appropriate. In the case of wild caught fish species, this approach is reasonably valid, as catches are constrained by management measures as well as stock availability. While there may be some potential for the production of aquaculture species to respond to price changes, the time taken to grow these species to a harvestable size limits this response over a short time horizon.

Several different functional forms of the model are commonly applied. These include a linear form of the model, a log linear form of the model, a Box-Cox transformation of the model, and a semi-logarithmic form of the model, respectively given by (1) $$P_s={\beta }_{0,s}+\sum _j{\beta }_j{X}_{j,s}+{\epsilon }_s;$$(1) (2) $$\hspace{0.17em}\ln \hspace{0.17em}(P_s)={\beta }_{0,s}+\sum _j{\beta }_j\hspace{0.17em}\ln \hspace{0.17em}(X_{j,s})+{\epsilon }_s;$$(2) (3) $$\left(P_s^{\lambda }-1\right)/\lambda ={\beta }_{0,s}+\sum _j{\beta }_j{X}_{j,s}+{\epsilon }_s,$$(3) and (4) $$\hspace{0.17em}\ln \hspace{0.17em}(P_s)={\beta }_{0,s}+\sum _j{\beta }_j{X}_{j,s}+{\epsilon }_s.$$(4) where $P_s$ is the price of species s, $X_{j,s}$ is the characteristic j of species s, and ${\epsilon }_s$ is a random error term. All four types of models were estimated, and there is no a priori reason to suggest that one model will be more appropriate than others. It is generally accepted that the appropriate econometric model and functional form should be chosen based on the specific application and the available data (Costanigro et al., 2011).

Market price data and fish characteristics

The study used a time series of price and quantity data compiled by Pascoe et al. (2021), covering the period January 2000 to September 2019. These included monthly data for key Australian wild caught species landed at the Sydney Fish Market, imported fish species and Australian farmed salmon.

The Sydney Fish Market is the largest market of its kind in the Southern Hemisphere, and is the third largest seafood market in terms of variety in the world (Hussain et al., 2017). Fish and seafood sold on the market is sourced from Queensland, New South Wales and Victoria, as well as imports from New Zealand and elsewhere. The market is a key component of the supply chain for many high valued farmed and wild caught species (Hobday et al., 2011). Previous studies have found that prices on the Sydney and Melbourne markets—supplying the two largest population centers in Australia (representing 60% of the urban population of eastern Australia)—were highly cointegrated (Hoshino et al., 2021), suggesting that prices on the Sydney Fish Market were representative of eastern Australian prices in general.¹

For this analysis, only data for the period January 2012 to September 2019 were used. While earlier data were available, key import species such as Nile Perch, Tilapia and Basa could only be separately identified in the import data after 2012 following of a re-classification of the import codes. More recent data were available but not sought. Several studies have identified substantial impacts of COVID-19 on fish prices and value chains globally during 2020 and 2021 (e.g. Abe et al., 2022; Akter, 2020; Plagányi et al., 2021). In Australia, divergence of previously exported products to the domestic market, loss of restaurant and café trade due to lockdowns and density restrictions, impacts of these restrictions on local production, and the physical closure of the markets in some months meant that price series over this period were likely to be distorted. Ex post analyses (e.g., Curtotti et al., 2022; Ogier et al., 2021) found that these impacts varied considerably for different seafood products. To avoid the potential distortions that such data may create, additional data after September 2019 were not sought. An a priori assumption of the analysis is that markets will return to pre-COVID behavior once the pandemic has passed.

The data used from the Sydney Fish Market include the most common white fish species landed, including a range of high, medium and lower valued species. The fish are sold at auction in a range of product forms, but two forms make up around 90% of the product - whole (81%) and gilled, gutted and headless (9%). Only these two product forms were used in the analysis, identified separately in the data. All domestic wild caught fish are sold fresh.

Imports were landed either fresh or frozen, and in a variety of product forms with the dominant forms being fillets or whole. These two forms were used in the analysis. Hake (as a generic term) is imported from multiple countries, and the classification of Hake covers several different species. Most Hake comes from South Africa and Namibia; South America (Chile, Argentina) and New Zealand, but is also imported from several other countries. To ensure consistency of product, Hake from South Africa and Namibia was grouped as “Hake,” and Hake from New Zealand was renamed “Hoki.” Imported aquaculture species (Basa, Nile Perch and Tilapia) were also identified as farmed. Only imports landed into New South Wales were considered as these were assumed to mostly supply the same market as, and potentially compete with, the Sydney Fish Market (Pascoe et al., 2022).

Farmed salmon in Australia is predominantly produced in Tasmania, from where it is distributed to the rest of the country. Per capita consumption of salmon has remained fairly constant over time (Curtotti et al., 2022), so the available quantity of salmon used in the analysis was based on the total production pro-rated on the basis of the relative population of New South Wales.

Weights of all species (imported and domestically produced) were converted to liveweight equivalent, and prices determined from the ratio of value to quantity (Pascoe et al., 2021). Prices therefore are also liveweight equivalent, with price differences representing the value added due to different product forms (rather than an artifact of different weights). All prices were converted to 2019–2020 values using the consumer price index and are expressed in Australian Dollars (AUD$).

Information on a total of 20 different species, several with different product forms, were included in the analysis. The species were selected to ensure a wide range of characteristics were covered, including high- and low-priced species, high and low volume species, imported and domestically produced species, farmed and wild caught species. Most species were also sold on the market in a range of different product form (e.g., fresh or frozen; whole, gutted and headed or filleted), with each treated as a separate category for the analysis. A summary of the price, origin and product form information for each of the species is given in the Supporting Information (S1).

Flavor and other characteristics for each species were largely derived from online marketing resources, such as fishfiles.com. The key characteristics for which consistent descriptors could be obtained were thickness of the fillet, color and texture of the flesh, flavor, oiliness, and moistness. These were classified into levels (e.g., thick, medium, thin) and treated as categorical factors in the model. Details on the key characteristics of each of the species is given in the Supporting Information (S2).

The quantity supplied to the market was also included in the model. Assuming inelastic (and exogenous) supply in any one time period (McConnell & Strand, 2000) and an inverse demand for seafood products, then the price paid will also fluctuate with the quantity supplied (Barten & Bettendorf, 1989). While not commonly used in hedonic price models, quantity has been applied in several previous studies as factor explaining variations in price between time periods (e.g., Hammarlund, 2015; Lee, 2014; Martínez-Garmendia, 2010; Sogn-Grundvåg et al., 2019). Wolff and Asche (2022) found that the size of the lot being auctioned (not just total quantity) also impacted the price received.

The potential for seasonality to influence price was considered through the inclusion of monthly dummy variables (with January the base), while potential time trends were considered through annual dummy variables (with 2019 the base).

Sensory testing

As T. anak currently does not have a market presence, sensory properties including texture and flavor in contrast to common white fleshed fish species had to be established to generate reliable input characteristics for the hedonic price model. Sensory testing was undertaken to provide information relevant to the model, as well as provide an indication as to which species it may be most closely related to from a consumer perspective. Sensory testing involves a panel of trained testers assessing samples of different fish species, prepared under strict protocols, and noting their characteristics against a range of pre-defined descriptors. Descriptive analysis has been successfully used to characterize sensory properties of fish species in multiple studies. For example, Frank et al. (2009) conducted descriptive analysis to compare barramundi across a set of wild and aqua-cultured fish, while Aubourg et al. (2002) used descriptive analysis to describe rancid odors of frozen horse mackerel. Therefore, descriptive analysis was selected as the preferred method for the sensory evaluation.

Sample selection and preparation

Ten white fleshed fish samples including T. anak underwent sensory evaluation. The other nine samples included Basa (Pangasianodon hypophthalmus), Hoki (Macruronus novaezelandiae), Flathead (Platycephalus richardsoni), Pink ling (Genypterus blacodes), Murray cod (Maccullochella peelii), farmed Kingfish (Seriola lalandi), wild-caught Kingfish, farmed Barramundi (Lates calcarifer) and wild-caught Barramundi. These were selected as they represent a mix of common fish sold in Australia (e.g., domestic species such as Flathead, Pink Ling and Barramundi, and imported species such as Hoki and Basa), as well as less common species (e.g., Kingfish and Murray Cod) which were potentially less readily identified by the panelists. Two of the species tested were sourced from both capture fisheries and aquaculture (i.e., Kingfish and Barramundi). Market prices of the species also ranged from low to high. Hence, it was possible to determine where T. anak would be placed within this group.

As all T. anak were supplied as frozen fillets, genetic testing was undertaken to ensure the correct species was included in the study and correctly identified. The frozen fish fillets were stored at −20 °C and all samples were thawed in a refrigerator at 4 °C overnight. They were then portioned into 50 g samples, wrapped in aluminum foil, and labeled with a 3-digit code. The prepared fishes were placed in a multicooker, single layered and steamed for 10 minutes. For food safety and consistency, random temperature checks were conducted on cooked samples to ensure they were ≥69 °C and visual inspections carried out to ensure that the samples were flaky before serving to the panelists. Steaming was chosen as the cooking method to minimize cooking loss, ensure uniform heating and to not alter or mask the aroma or flavor of the fish samples (Alasalvar et al., 1997; Huang et al., 2021).

Sensory analysis

The testing was performed by ten experienced and trained panelists who were previously screened for sensory acuity (Lawless & Heymann, 2010). Assessors participated in five 2-hour training sessions where, through multiple exposure to the samples and moderated discussion, they were introduced to vocabulary generated from literature review. They were asked to generate additional descriptive vocabulary if required and developed a consensus vocabulary using Check-All-That-Apply (CATA) methodology and a corresponding method of assessment that best described the sensory differences (odor, texture, taste and flavor, and after feel/aftertaste) between the products. A range of reference standards were sourced in the supermarket as comparison and to improve agreement on some of the odor and taste/flavor attributes included in the descriptive vocabulary (see Supporting Information, Table S3).

Table 3. Estimated price of farmed and wild caught T. anak (liveweight equivalent, $kg).

	Log-linear		Linear		Box-Cox		Semi-log
	Price ($/kg)	SE	Price ($/kg)	SE	Price ($/kg)	SE	Price ($/kg)	SE
Large (130 cm)
100 tonnes/year	10.06	0.76	11.46	0.47	10.05	0.64	8.58	0.69
500 tonnes/year	9.13	0.67	11.28	0.48	9.83	0.63	8.37	0.68
2,000 tonnes/year	8.40	0.62	10.60	0.51	9.02	0.64	7.62	0.66
Medium (65 cm)
100 tonnes/year	10.43	0.86	8.56	0.44	8.04	0.53	7.58	0.59
500 tonnes/year	9.48	0.77	8.38	0.46	7.85	0.54	7.40	0.59
2,000 tonnes/year	8.71	0.71	7.70	0.54	7.16	0.59	6.73	0.62
Small (30 cm)
100 tonnes/year	10.11	0.70	8.93	0.29	8.68	0.40	8.50	0.50
500 tonnes/year	9.18	0.61	8.75	0.30	8.48	0.40	8.30	0.49
2,000 tonnes/year	8.44	0.56	8.07	0.36	7.75	0.43	7.55	0.51
Wild caught (130 cm)
1 tonne/year	14.82	1.24	13.59	0.44	13.15	0.67	12.18	0.85
5 tonnes/year	13.45	1.02	13.59	0.43	13.14	0.67	12.18	0.85

Evaluation was conducted in duplicate over two sessions in March 2022. Assessors evaluated all ten samples each session. Each sample, identified only by a 3-digit code, was presented to assessors one at a time and in randomized order. Samples were served within one minute of preparation, wrapped in aluminum foil. A two-minute break between each sample and a five-minute break after five samples in the sample set was enforced. Crackers and warm water were used as palate cleansers between samples.

Assessors were seated in individual sensory booths that conformed to ISO 8589:2007 standards and samples under red lighting to mask differences in appearance. The intensity of each sensory attribute was rated on an unstructured line scale anchored at low (5%) and high (95%). Assessors were not required to provide an overall preference ranking for the species tested as a whole. Data were collected electronically using RedJade sensory software. Informed consent was obtained from panelists and this study was approved by the CSIRO Human Research Ethics Committee (approval number: 2021_050_LR sub application C).

Statistical analysis

The results were analyzed using SenPAQ (version 6.3) software. The differences between samples were investigated using 2-way ANOVA with samples (n = 10) as fixed factor and participants (n = 10) and their interactions as random factor for each sensory attribute. Least Significant Difference (LSD) post-hoc analysis was conducted to determine if there were any significant differences between samples at a 95% confidence level. To visualize the differences in sensory characteristics between fish samples a canonical variate analysis (CVA) was performed using XLStat (version 2019.3.1).

Results

Regression analysis

The four different functional forms of the models, given by Equations (1)–(4), were estimated with price of each species (whole weight equivalent) as the dependent variable (logged, unlogged or Box–Cox transformed as appropriate), and the characteristics of the different species and products (detailed in the Supporting Information S1 and S2) as the explanatory variables. The key characteristics were included as categorical variables (factors), while product form, and origin (i.e., imports) were included as dummy variables. Quantity supplied and fish length were included as continuous variables. An interaction term between imports and quantity supplied was also included to allow for the potential for different price flexibilities for imported goods relative to domestically produced seafood. Time and month dummy variables were also included to capture any potential changes over time and due to seasonality. The models were estimated as random effects panel data models.²

The potential for multicollinearity in the models was tested based on the generalized variance inflation factor (GVIF). While multicollinearity does not affect the predictive power of a model, it can result in distorted estimates of the individual parameter values. The estimated GVIF are provided in the Supporting Information (Table S5). From these, multicollinearity was not considered a problem in the models.

The estimated coefficients of the four different regression models are summarized in Table 1. The time related coefficients are not included in Table 1, but the full models are presented in Tables S6–S9 in the Supporting Information. All models performed well in terms of goodness of fit (${\overline{R}}^2$), explaining between 79% and 83% in the variation in the dependent variable. As the model had different dependent variables (due to the different transformations of price as given in Equations (1)–(4)), a direct comparison of ${\overline{R}}^2$ is not appropriate as a criterion to assess which model performed the best. Estimates of the Akaike Information Criteria (AIC) and the Bayesian Information Criteria (BIC), following Banks and Joyner (2017), suggests that the log-linear model is the most appropriate model. A Box-Cox test was also undertaken (see Table S10 in the Supporting Information), confirming that the log-linear model is the most appropriate functional form.

Table 1. Regression models coefficients with clustered and robust standard errors.

	Log-linear model			Linear model			Box–Cox model			Semi-log model
	${\beta }_j$	Std. error	Sig.	${\beta }_j$	Std. error	Sig.	${\beta }_j$	Std. error	Sig.	${\beta }_j$	Std. error	Sig.
Intercept (${\beta }_0$)	1.975	0.124	***	5.109	0.156	***	1.977	0.046	***	1.537	0.030	***
Continuous variables^a
Quantity	−0.060	0.007	***	0.000	0.000	***	0.000	0.000	***	0.000	0.000	***
Length	0.041	0.028		−0.011	0.006	+	−0.004	0.002	**	−0.003	0.001	***
Quantity*Import	−0.091	0.009	***	0.000	0.000	***	0.000	0.000	***	0.000	0.000	***
Dummy variables
Import	0.295	0.078	***	−4.014	0.307	***	−0.860	0.093	***	−0.439	0.061	***
Fillets	−0.337	0.024	***	−2.320	0.126	***	−0.908	0.042	***	−0.645	0.029	***
Gilled and headed	0.055	0.021	**	0.066	0.098		0.054	0.029	+	0.040	0.020	*
Frozen	0.142	0.038	***	−1.216	0.184	***	−0.218	0.065	**	−0.078	0.045	+
Farmed	−0.108	0.049	*	−2.082	0.361	***	−0.545	0.102	***	−0.344	0.067	***
Factors
Thickness: Thin	–			–			–
Thickness: Medium	−2.160	0.085	***	−11.360	0.657	***	−3.255	0.166	***	−2.008	0.100	***
Thickness: Thick	−0.065	0.032	*	3.604	0.515	***	0.703	0.133	***	0.336	0.078	***
Flesh: Off-White	–			–			–
Flesh: Pink	0.722	0.029	***	2.225	0.187	***	0.719	0.056	***	0.483	0.039	***
Flesh: White	−0.554	0.081	***	−6.599	0.758	***	−1.499	0.194	***	−0.813	0.117	***
Texture: Soft	–			–			–
Texture: Medium	0.259	0.035	***	3.131	0.202	***	0.763	0.056	***	0.426	0.037	***
Texture: Firm	0.855	0.056	***	6.272	0.359	***	1.742	0.100	***	1.053	0.066	***
Flavor: Mild	–			–			–
Flavor: Medium	0.133	0.040	**	4.393	0.785	***	1.040	0.205	***	0.592	0.121	***
Flavor: Sweet	0.914	0.063	***	9.606	0.854	***	2.281	0.220	***	1.270	0.130	***
Oiliness: Low	–			–			–
Oiliness: Medium	−0.714	0.046	***	−3.944	0.399	***	−1.207	0.104	***	−0.763	0.063	***
Oiliness: High	0.478	0.056	***	6.528	0.714	***	1.598	0.193	***	0.914	0.117	***
Moisture: Moist	–			–			–
Moisture: Medium	−0.478	0.050	***	−8.647	0.773	***	−1.961	0.202	***	−1.070	0.121	***
Moisture: Dry	−1.109	0.047	***	−13.139	0.928	***	−3.061	0.236	***	−1.707	0.138	***
Diagnostics
RSS	259.78			7,193.60			693.73			315.71
Scaled RSS	259.78			414.97			294.79			315.71
${\chi }^2$ (DF = 38)	12,067.2		***	12,315.8		***	10,774.7		***	9,405.8		***
${\overline{R}}^2$	0.829			0.833			0.814			0.792
Hausman ${\chi }^2$ (DF = 20)	6.229			20.754			14.534			10.628
AIC	−5,488.06			2,713.12			−2,914.19			−4,808.32
BIC	−5,261.38			2,938.77			−2,688.55			−4,582.68

Significance codes: 0 “^***” 0.001 “^**” 0.01 “^*” 0.05 “⁺” 0.1 “” 1. ^aContinuous variables were logged in the log-linear model, otherwise unlogged. The value of lambda for the Box-Cox model was estimated to be 0.3 (see Figure S1 in the Supporting Information).

Most of the estimated coefficients were found to be statistically significant from zero. In most cases, the sign of the coefficients were the same across the different models. The values of the coefficients, however, are not directly comparable as they relate to different transformations of the dependent variables. Robust and clustered standard errors were estimated to correct for the potential bias in the standard error due to the clustered nature of the data (Abadie et al., 2022; Colin & Miller, 2015), clustered at the Year-Month level. With the exception of the time-related variables, the coefficients were largely significant individually. The model coefficients were jointly significant as indicated by the ${\chi }^2$ statistic, suggesting that the models may provide reasonable price predictions.

Fedderke and Li (2020) suggest that an appropriate test for the robustness of hedonic price models is to test the models against known outcomes. The models were tested against the known average prices of a range of species and product forms. The species selected (Table 2) included both farmed and wild caught species; high, medium and low valued species; and imported and domestically produced species. The delta method was used to derive the standard errors of the price estimates.

Table 2. Comparison of model estimated price with average market price from the data (live-weight equivalent, $/kg^a).

Species and product form	Market price	Log-linear		Linear		Box-Cox		Semi-log
		Estimate	SE	Estimate	SE	Estimate	SE	Estimate	SE
	($/kg)	($/kg)		($/kg)		($/kg)		($/kg)
Basa (frozen fillets)	1.42	2.44	0.15	5.12	0.39	3.29	0.25	2.72	0.20
Hoki (frozen fillets)	1.62	1.28	0.04	0.20	0.17	1.25	0.06	1.39	0.05
Silver wahoo (whole)	1.82	2.25	0.08	3.09	0.15	2.59	0.10	2.43	0.09
Tilapia (frozen whole)	2.70	2.79	0.09	4.32	0.16	4.04	0.13	3.99	0.13
Tilapia (fresh fillets)	2.96	2.33	0.11	3.19	0.23	2.50	0.15	2.29	0.12
Gemfish (GH)	3.75	3.51	0.12	3.76	0.18	3.69	0.14	3.64	0.14
Eastern school whiting (whole)	4.03	4.30	0.11	4.73	0.15	4.37	0.12	4.27	0.12
Morwong (whole)	4.67	4.52	0.19	4.90	0.25	4.74	0.21	4.69	0.21
Blue wahoo (GH)	4.72	2.49	0.10	3.17	0.17	2.70	0.12	2.54	0.11
Silver trevally (whole)	6.30	6.35	0.21	6.97	0.24	6.53	0.25	6.35	0.25
Ling (GH)	7.08	7.73	0.19	8.05	0.16	7.93	0.18	7.88	0.21
Flathead (whole)	7.95	8.50	0.19	8.55	0.15	8.42	0.17	8.35	0.20
John dory (whole)	12.57	9.09	0.26	10.78	0.21	10.16	0.25	9.81	0.29
Blue-eye trevalla (GH)	13.05	11.66	0.28	12.65	0.19	12.24	0.25	11.98	0.32
Salmon (whole)	14.50	15.08	0.31	14.86	0.24	14.94	0.41	14.93	0.54
Average absolute error (%)		18%		40%		24%		21%
Maximum error (%)		72%		261%		132%		92%

Notes: GH: Gutted and Headed. ^aUnless otherwise specified, all prices are in Australian Dollars (AUD), where 1AUD = USD0.69 in July 2022.

From Table 2 and Figure 1, all models provided reasonable estimates for the mid-range price species (e.g., Morwong), but performed poorly against John Dory (error ranged from −14% to −28%), a higher valued wild caught species and Blue Wahoo (error ranged from −33% to −47%), a mid-range valued species. Both species are characterized by relatively low volumes. Prices at the lower end of the market were also less consistent. All models overestimated the price of Basa (error ranged from 72% to 261%) but underestimated the price of Hoki (error ranged from −14% to −87%) (Table 2).

Figure 1. Comparison of observed and estimated prices of selected species.

Sensory experiments

Overall, the ANOVA results of the descriptive sensory analysis show that there were significant differences between the fish samples in terms of odor, texture, taste and flavor, and aftertaste/afterfeel. The ANOVA results, sample means for each of the sensory attributes and post-hoc comparison results are summarized in Supporting Information (Tables S4). The dimension 1 (F1) of CVA biplot (Figure 2) represented 59.22% of the data separating the samples based on texture, while dimension 2 (F2) accounted for 17.32% differentiating samples on aroma and flavor intensity, explaining a total of 76.54% of data variability. This biplot indicates that Basa and Flathead were perceived very different to other fish in terms of sensory properties, by being high in odor intensity while being opposed to each other in texture characteristics, with Basa being described as tender and juice and Flathead as firmer and chewier.

Figure 2. (a) CVA plot of all ten fish samples and (b) sensory attributes.

The remaining fish were separated based primarily on their textural properties along the horizontal axis. Panelists could not perceive a difference between farmed and wild-caught kingfish since they are overlapped. T. anak showed sensory characteristics similar to Pink Ling and Kingfish, and was perceived as a fish with firm texture; low to neutral odors and flavors such as ammonia; and low oily mouthcoating. In terms of texture, T. anak was also perceived as a firmer fish compared to Barramundi, which was considered to be a fish with tender flesh.

Not all factors used in the regression model were covered in the sensory analysis. For example, fillet thickness was not considered as a sensory attribute, as each of the samples tested were provided in similar proportions. However, given the overall similarity between T. anak and Pink Ling, some of the “missing” attributes could be assumed, or derived from other sources. For the purposes of estimating the potential price of T. anak, it was assumed that flesh color (not featured in the sensory analysis) was an off-white color (based on photographs taken during sample preparation and on other Dart species sold on the Sydney Fish Market³) and commercial fillet thickness was assumed to be related to the fish size, and assumed to be thick if fully grown fish were supplied to the market. The maximum size of the species was assumed to be 130 cm.⁴ From the sensory analysis, the texture was considered firm, flavor considered mild, oiliness considered low and flesh moisture was considered to be moist.

Price estimates for T. anak

The potential price of T. anak given the information derived from the sensory analysis was estimated using all four models (Table 3). The model used information on the characteristics from the sensory testing and other sources, and considered different levels of production and three different harvest size: fully grown (130 cm), medium sized (65 cm) and small (30 cm); and associated assumed thicknesses: thick, thin, thin (given the smaller sizes). The choice of fish size was arbitrary for the purposes of illustration only, although it is expected that aquaculture production will most likely focus on the smaller to medium end of the size range. For comparison, the potential price of (fully grown) wild caught T. anak was also estimated. Again, the delta method was used to derive the standard errors of the price estimates.

The estimated fish price decreased as the volume of production increased in all models, as would be expected (given the economic law of demand). The Linear model resulted in the highest estimates of fish price for the farmed species, although the Log-linear model estimated the highest price for the wild-caught species. The Semi-log model consistently estimated the lowest price for the species.

Producers have different options in terms of grow out of the species, potentially producing fewer larger fish or a greater density of smaller fish, with potentially more rotations as well. Associated with longer grow-out is also a greater feed cost and increased risk. The Log-linear model suggests that the price received for smaller fish is less (per kg) than that of the larger fish, while the trend was more complex in the other models. From the regression model (Table 1), prices in these models decreased with fish length. However, fish with thicker fillets (as larger fish would have) also a price premium over those with thinner fillets. This resulted in prices initially decreasing as the fish size decreased (due to moving to thinner fillets), but then increasing as size decreased further. Non-linear outcomes with size have been observed in other studies. For example, some studies have also found that price tends to increase with size (e.g. Tsikliras & Polymeros, 2014), while others suggest that smaller “plate-sized” fish have a price premium over larger fish (Reddy et al., 2013). Whichever is the more appropriate assumption in this case, the estimated price “premium” in either direction is relatively small. This suggests that producing smaller sized fish may be more profitable (saving feed costs) regardless of which model is correct.

The difference between the estimated wild caught and farmed price can be attributable in part to the lower volume of wild-caught fish supplied to the market (based on observed commercial catches over the last 20 years) as well as the estimated aquaculture “penalty” in the regression models (Table 1). For example, from Table 1, the coefficient on farmed product in the Log-linear model suggests that prices for aquaculture produced species are 90% (i.e., exp(−0.108)) of their wild caught equivalent, ceteris paribus. Lower prices for farmed species have also been seen elsewhere, with consumers generally having a higher willingness to pay for wild caught species than their farmed counterpart (e.g. Davidson et al., 2012), as well as a general perception of higher quality of wild-caught fish (e.g. Claret et al., 2014). In contrast, Ray et al. (2022) found that the impact of production method varied by species, with some species receiving a higher price for farmed fish and others for wild caught.

Discussion

A key variable in the assessment of the financial viability of a new aquaculture enterprise is the price received, and how this varies by size (which affects rotation and/or grow out length) and production level. In the case of T. anak, this information was not available, and the potential market price of the fish species needed to be estimated.

The hedonic price modeling approach adopted to derive these estimates is more commonly used to estimate the contribution of different characteristics to the overall price of a good. In fisheries, as noted earlier, this has included the value of sustainability certification (Asche et al., 2021) and other attributes such as fish size, freshness and product form on price (e.g. Hammarlund, 2015; Hukom et al., 2020; Kristofersson & Rickertsen, 2007; Lee, 2014; McConnell & Strand, 2000; Pettersen & Asche, 2020; Ray et al., 2022; Roheim et al., 2007). However, hedonic price models have been used to derive expected prices given a set of characteristics in other contexts. For example, Malaitham et al. (2020) used the approach to estimate fair compensation prices for property to be resumed for transportation infrastructure development, while Fedderke and Li (2020) used the approach to estimate guide prices for art auctions. In this regard, the use of the approach to estimate the price of a species currently not common on the market is reasonable.

The analysis was limited by the available data. Although a wide range of species was considered, the data did not contain any tropical species due to lack of appropriate data. Tropical species such as coral trout (Plectropomus sp.), red throat emperor (Lethrinus miniatus) and tropical snappers (Lutjanus sp.) are generally high value species. Given that T. anak is also a tropical species, it may share characteristics more common to these species than the species considered. Information on prices at an annual (rather than monthly) level for most of these species is available (e.g., Steven et al., 2021), but reliability of these data is uncertain. For example, the reported price of coral trout and red throat emperor (i.e., total value/total quantity) has remained constant in real terms since 2005–2006 (derived from Steven et al., 2021). Similarly, a reliable time series of data were not available for barramundi (Lates calcarifer), which is both farmed and wild-caught.

While the commercial catches of T. anak are currently small (1–5 tonnes a year), the expectation of the aquaculture activities is to potentially produce 500–1,000 tonnes a year. This would make it comparable in terms of volume with many of the wild caught species landed in Australia and used in the analysis. In this regard, the set of species used in the analysis are considered reasonable for the purposes of comparison with the planned production of T. anak.

Descriptive sensory analysis provided valuable information for the price modeling, as this species currently does not have a market presence and no reliable information of its sensory characteristics required for the model were available in the literature. T. anak was described as being of firm texture and mild in flavor, sharing characteristics with well-regarded dining species such as Kingfish and Pink Ling supporting the potential of T. anak as a new aquaculture species. This characterization aligns with anecdotal descriptions of other Trachinotus species being a fish with firm white flesh with a delicate and palatable taste; however, further research is required to reliably characterize and compare different Trachinotus species. T. anak also lies opposite Barramundi (Figure 2), one of the most common white fleshed fish in Australia in terms of texture, which shows the potential of T. anak to compete in the market without necessarily displacing consumption of another Australian aquaculture fish.

The analysis excluded interactions between variables, other than quantity and the import dummy variable. Asche et al. (2021) and Ray et al. (2022) found that the effect of product form and other characteristics on price may vary between species, effectively defining sub-markets for each. As a result, modeling species separately may provide a better estimate of the contributions of their characteristics to their price. While this may also be the case also for the Australian and imported species considered in our models, the aim of the study was to provide an estimate of prices received for a species currently not on the market. As a result, our models provide more of an estimate of the average effects of the characteristics across species (rather than a more accurate species-specific effects). Given that the models were able to provide reasonable estimates of most of the species against which they were tested, extrapolating these average effects to T. anak is likely to be reasonable.

The model is derived from observed auction-based market prices, and in some cases includes value-adding through processing. The latter is considered directly in the model through the conversion to live-weight equivalents and also includes dummy variables to capture the effect of product form on price. On-farm processing of T. anak could result in higher prices being realized to the farmer than the live-weight equivalent values indicated above, although would also result in additional costs. While the models suggested that filleted fish resulted in a reduction in the live-weight equivalent price, this needs to be traded-off with other considerations such as access to other markets (e.g., restaurants, catering etc), which may yield higher prices, processing costs, and potentially lower transportation and marketing costs. These costs will also vary from location to location (i.e., distance to the market, number of agents, transportation options, packaging needs, auction or wholesale market options, etc.). These additional costs have not been taken into consideration in this analysis, and will impact the overall profitability of the producer.

The estimated prices of T. anak are roughly similar in magnitude to prices received for related species in other markets. For example, Weirich et al. (2021) suggested that the farm-gate price for Florida pompano (T. carolinus) was around USD10/kg in 2019, roughly equivalent to AUD14.50/kg (live-weight equivalent), while Santhiya et al. (2021) suggested a farm-gate price of USD8/kg in 2019 (AUD11.53) for the same species. In contrast, Lan et al. (2022) suggested that farm gate prices for T. anak in Taiwan ranged from NTD150 to NTD165 in 2019, roughly equivalent to AUD7.22 to AUD7.94. These markets are substantially different to the Australian market, and in the case of the Florida pompano, the fish characteristics may also differ, so a direct comparison is difficult. It is somewhat reassuring, however, that the price estimates from the hedonic price model fall within the range of these observed prices elsewhere.

The introduction of a “new” species into the market may also require additional marketing to achieve the prices estimated by the model. Consumers unfamiliar with the species may not be initially willing to pay the price that its characteristics might otherwise suggests. Evidence elsewhere is that consumer learning is slow, and that prices of new products are generally lower than their ultimate (equilibrium) level (Caminal & Vives, 1999). Effective marketing strategies can improve the speed at which this equilibrium is reached (Durmusoglu et al., 2022), and the development of such a strategy to promote the product will be beneficial.

Prices are only one of the factors affecting the economic and financial feasibility of the development of a new aquaculture industry; production costs are also an important consideration (Asche, 2008). The development of farm-level bioeconomic models is required to combine costs and prices with biological information to assess the potential feasibility of the industry as well as identify potential areas where costs saving may be realized. For example, the potential for selective breeding to reduce feed costs through higher feed conversion ratios (de Verdal et al., 2018), increase growth (Gjedrem & Robinson, 2014), increase fillet yield (Vandeputte et al., 2019) or reduce mortality (Berrill et al., 2012) can also enhance the financial and economic feasibility of the industry.

Conclusions

Liao and Huang (2000) noted eight key characteristics of potential successful aquaculture species, the first two being a fast growth rate and a high economic value, although the most successful in terms of quantity produced were those species that realized significant productivity growth, reducing both cost and price (Asche, 2008). Most current aquaculture species satisfy these conditions, although some species trade off high growth against low economic value. For species that do not have an existing market profile, however, assessing the potential economic value is difficult.

Hedonic price models provide a means to derive estimates of prices for “new” species. In this case, we have used four different functional forms of the hedonic price model to consider uncertainty due to model choice as well as due to estimated parameter uncertainty in each of the models. In terms of validation, the model parameters across each of the models generally had the same signs and were consistent with a priori expectations based on economic theory (i.e., prices declined as quantity increased). The estimated prices derived from the model were also largely consistent with those of known species.

The study also demonstrates the importance of sensory evaluation of different fish species for estimating the potential market price. Hedonic pricing assumes that the price of the fish is a function of its eating characteristics. While these are more or less known for existing species on the market, for “new” species these characteristics need to be determined. Sensory evaluation provides an established and robust process to derive these key characteristics.

The results of the analysis suggest that the price that is likely to be received for T. anak if aquacultural production is to proceed is most likely to fall in the range from $8.71/kg to $10.43/kg for medium sized fish (around 4 kg) and $8.44/kg to $10.11 for small fish (around 1 kg), depending on the quantity produced and based on the results from the most appropriate model (i.e., Log-linear). The size of the cultured fish at market reflects consumer demand while also capitalizing on the fish’s exponential growth phase i.e., as the fish gets larger the growth slows. Given the cost of production increases with size, it is unlikely that larger fish (>6 kg) will be produced, and hence these values (adjusted also to account for marketing costs) are of more relevance to potential producers. These prices may not be achieved immediately due to consumer unfamiliarity with the product, but combined with an effective marketing strategy, should be achieved once consumer familiarity with the product is obtained.

Acknowledgements

The authors would like to thank the Editor and the two anonymous reviewers for their useful comments on earlier drafts of the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Brisbane, the capital of Queensland and the third largest city in Australia, does not have a central seafood market.

2 A Hausman test (Hausman, 1978) was undertaken to determine if random or fixed effects was the most appropriate assumption. In all cases, the models with the different effects specification were not significantly different, but as random effects are more efficient these were used in the final analysis.

3 https://www.sydneyfishmarket.com.au/Home/Seafood/Species-Information/List/dart

4 https://reeflifesurvey.com/species/trachinotus-anak/; https://fishesofaustralia.net.au/home/species/1664