Fishing vessel efficiency, skipper skills and hake price transmission in a small island economy

Abstract

Determinants of vessel efficiency and vertical/horizontal price transmissions in consumer markets are key elements for assessing the viability of a fishery, particularly for a small fishery-dependent economy. An open issue also concerns whether vessel efficiency levels influence export prices. The paper sets off with a review of evidence from other countries, followed by hypotheses for the Falkland Islands. To test these hypotheses, the analysis first applies a stochastic frontier model accounting for latent skipper skills, to a monthly 2008–2016 panel of fishing vessels operating in the Islands. Using estimated vessel inefficiency by licence type as a proxy indicator of product quality and extra costs of transhipment, the study moves on to examine price adjustments of Falkland hake and other finfish sold at Spanish ports vis-à-vis two major south Atlantic hake supplier countries—Argentina and Namibia—and local traders. Lastly, based on full sample and rolling widow regressions on 2004–2016 monthly data, the analysis formulates and estimates threshold autoregressive models for the hake value chain in Spain, as the largest European port-of-entry and market for fresh and frozen hake, including from the Falklands. Once different output frontiers are accounted for, vessels with licences for hake as their main target do not outperform, in terms of technical efficiency, less-valued finfish vessels. Besides evidence of increasing integration within supplier and consumer markets, econometric results suggest some degree of price ‘leadership’ by Namibian hake exporters and asymmetric behaviour in short-run price adjustments by Spanish retailers. However, producer and consumer markets turn out to be weakly interlinked.

Appendices

Appendix 1: Hake and squid species in Falkland waters

Oceanographic conditions highly influence fish migratory flows and other seasonal features of the biological cycle in the Patagonian Shelf. As observed for other fisheries (Guijarro et al. 2012), reproductive and migratory habits induce intra-year patterns in population dynamics. Both southern hake species (Merluccius australis and Merluccius hubbsi) are seasonal migrants, moving from inshore spawning grounds, mainly in Argentina’s Exclusive Economic Zone (EEZ), to adult feeding grounds in Falkland seawaters. M. australis is mostly present in the western part of Falkland EEZ during the first half of the calendar year, in the austral summer and autumn, with relatively high catches in February–May. M. hubbsi tends to migrate a few months after, consequently starting their feeding season later (Falkland Islands Government (FIG) 2014; Portela et al. 2002).

The Patagonian shortfin squid (Illex argentinus) constitutes one of the most important fish resources in the Shelf. It has a life cycle of nearly 1 year with strong variation in biomass from year to year, and its distribution is limited to the area of confluence of cold and warm currents of sub-Antarctic (Falkland Current) and sub-tropical origin. Along with another local squid (Loligo gahi), Illex is among prey fishes for hakes, although the impact on stock appears to be limited since most predation mortality concerns fish of young age. The reverse also occurs, with young hake being preyed on by maturing Patagonian shortfin squids (Villasante et al. 2015). In years with no strong sea-surface temperature anomalies, the Falkland EEZ and adjoining high seas register peak concentrations of Illex between March and May. Squid migrations to warmer Argentine spawning grounds take place in July and August (Portela et al. 2005). At the end of the Illex fishing season, more frequent stormy weather conditions often hamper effective fishing, for a relatively small number of jigging vessels allowed to fish Illex till mid-June (Falkland Islands Government (FIG) 2014).

During the Illex season, vessels with G-licences, which includes Illex as a target species, often register the largest shares of hake catches, even if these account sometimes for less than 10% of total catches by G-licence fleet (Falkland Islands Government (FIG) 2014). The pattern of hake catches per fishing hour partly reflects these features of Falkland fisheries, even if a few vessels might have underreported the fishing time in months of high catches (Fig. 5: CPUE in tens of kilogrammes of hake per hour spent at sea, by month). Hake is mainly harvested in northern and western parts of the Falkland EEZ. This also applies to Illex and rockcod, with more dispersed distribution for the latter. Since its establishment in 1990, restricted grounds for fishing of Loligo gahi, which extend on the shelf edge around the Islands from east to south, have served a double purpose: keeping finfish vessels out of the Loligo squid fishery and avoiding incidental capture of juvenile finfish by vessels targeting Loligo (due to different minimum mesh sizes; Falkland Islands Government (FIG) 2014). Beyond the Falkland EEZ, a few pelagic trawler vessels are allowed to fish in South Georgia island seawaters, with icefish licences awarded on an annual—prior to 2014—and, more recently, biennial basis, according to stringent standards (Pelembe 2014).

Fig. 5

CPUE (Falkland hake, January 2008–July 2017)

Appendix 2: TAR models

Based on an observable regime-change signal (threshold variable) q_t, a TAR(m;p) model with autoregressive order p partitions strictly increasing values of a random variable into m regimes separated by m−1 threshold values γ_j (j = 1,…, m−1), with theoretically unbounded extremes γ₀ = −∞, γ_m = ∞. The model yields m regime-specific parameters. In particular, a TAR(2;1) is applicable to differenced residuals Δε_t, where ε_t−1 represents the error correction term from a cointegrating regression (Table 5: z_t−1). In general notation as TAR(m;1), this can be expressed as Eq. (8a) (testing APT in short-run readjustments from prior rates of change of deviations from equilibrium) or Eq. (8b) (testing APT in magnitude and strength of response to these deviations, that is cointegration with threshold non-linearity). The variance of the random error η_t may vary across regimes and 1_j(q_t, γ) is a Heaviside indicator function (= 1 if γ_j < q_t <γ_j +1; = 0 otherwise). Thresholds and parameter estimates are global minimisers of the objective function Eq. (9), where S_γ is the sum of squared residuals (η_t²) in the partitioned sample. In this equation, the ratios λ_[j] (= γ_j/γ) are such that each threshold value is distinct and bounded away from extreme values of q_t.

$$ \Delta {\varepsilon}_t={\Sigma}_{j=1,..,m-1}{1}_j\left({q}_t,\gamma \right)\cdot \left[{\varphi}_{0j}+{\varphi}_{1j}\Delta {\varepsilon}_{t-1}\right]+{\eta}_t\kern1.25em \left({\eta}_t\sim N\left[0,{\sigma_{\eta \mid j}}^2\right]\right) $$

(8a)

$$ \Delta {\varepsilon}_t={\Sigma}_{j=1,..,m-1}{1}_j\left({q}_t,\gamma \right)\cdot \left[{\alpha}_j{\varepsilon}_{t-1}\right]+{\varphi}_1\Delta {\varepsilon}_{t-1}+{\eta}_t\kern0.5em \left({\eta}_t\sim N\left[0,{\sigma_{\eta \mid j}}^2\right]\right) $$

(8b)

$$ \left({\gamma}_1,{\gamma}_2,\dots \kern0.5em ,{\gamma}_{m-1}\right)={\mathrm{argmin}}_{\left(\lambda \left[ 1\right],\lambda \left[ 2\right],\dots \kern0.5em ,\lambda \left[m-1\right]\right)}{S}_{\gamma}\left({\gamma}_1,{\gamma}_2,\dots \kern0.5em ,{\gamma}_{m-1}\right) $$

(9)

$$ {\Lambda}_{\tau }=\left\{\left({\lambda}_1,{\lambda}_2,\dots \kern0.5em ,{\lambda}_{m-1}\right);\left|{\lambda}_{j+1}-{\lambda}_j\right|\ge \tau, {\lambda}_1\ge \tau, {\lambda}_{m-1}\le 1\hbox{--} \tau \right\} $$

(10)

Bai and Perron (2003a/b) propose a dynamic programming algorithm solution of Eq. (9), subject to asymptotic consistency rules defined in Eq. (10) (where a trimming parameter τ is the ratio of a minimum segment length h to the range of q_t, i.e. τ = h/γ). Relative to an analogous treatment with breakpoints, which replaces γ_j with T_j and γ with T, see Bai And Perron (1998, 2003a). Threshold regression and breakpoint testing are fundamentally equivalent: in the latter, by permuting the observation index, time is the threshold variable (IHS 2015, p. 428; Tsay 1989).