Abstract
A vast literature in fisheries economics focuses on drivers of fishers’ behavior with limited attention given to what happens once the fish are landed. This often strongly contrasts with a main policy focus on coastal communities, with fisheries management as an additional instrument in supporting livelihoods. This study shows that the number of Norwegian landing plants has been reduced in recent decades, and that quantity landed, annual plant operation time, and attracting smaller vessels decrease the probability of exit. Interestingly, plants in communities with additional landing locations have lower probabilities of exit, pointing to an industry cluster effect. (JEL Q22)
1. Introduction
There is a vast literature in fisheries economics focusing on what happens on the water and on drivers of fishers’ behavior, while limited attention has been given to what happens after the fish is brought aboard a vessel. This is somewhat surprising as in many countries, developed as well as developing, the main policy focus is on coastal communities rather than on fisheries, with fisheries management often playing the role of an additional instrument in supporting livelihoods in these communities (Olson 2005). Fishers and coastal communities are frequently shown to be poor or in decline, and interventions through improved fisheries management are surprisingly often portrayed as detrimental (Munk-Madsen 1998; Copes and Charles 2004; Olson 2006; Hersoug 2011; Tietze 2016).1 As a consequence, fisheries management systems tend to build in features that are meant to enhance the social objectives of the fishery, even though they may be economically inefficient (Kroetz, Sanchirico, and Lew 2015). Subsidizing fisheries (Sumaila et al. 2016), allocating individual processing quotas (Matulich and Sever 1999), community development quotas (NMFS-ARO 2017; NPFMC 2017), and imposing obligations to land fish in specific communities (Standal and Aarset 2008; Hermansen and Dreyer 2010) are some examples of management tools meant to help fishery-dependent communities.
When landing their catch, fishers must choose between different landing plants, invariably associated with coastal communities due to their geographical location. A landing plant can be defined as a connecting node in the fish supply chain, where catches are landed. At some plants the fish may be further processed, while at others they will just be distributed to the next level in the supply chain. These nodes are also the main link to the coastal communities that are supported by the fishing activity. In struggling coastal communities, at least some of these plants are performing poorly from an economic perspective, and worst case, the coastal community’s link to the fishing industry is weakened or interrupted when the last plant closes down. Hence, for coastal policy and communities it is of substantial interest to investigate if there are any factors that help explain which landing plants thrive and which do not.
This paper investigates attributes that may influence landing plants’ survival in the Norwegian groundfish sector. The methodology applied is similar to the more general literature for firm survival (e.g., Salvanes and Tveteras 2004). Similar econometric approaches are also used to investigate choice of fishing location (Bockstael and Opaluch 1983; Eales and Wilen 1986; Smith and Wilen 2003; Huang and Smith 2014) and choice of fishing gear (Eggert and Tveteras 2004). A pooled logit model and conditional logit models are used to estimate the probability of a plant be-coming inactive, in association with a range of attributes. Cox proportional hazard models are applied to identify factors that affect the duration of a landing plant’s activity and to estimate the probability of its survival. The analysis is carried out on a dataset containing all groundfish landed in the northern half of Norway, an area comprising the main fishing regions in the country.
2. Background and Data
The landing plants that receive groundfish in Norway are spread out along the coast, with some concentration in areas close to popular fishing grounds, such as for cod among the Lofoten islands. Groundfish is the most important fishing sector in Norway, with approximately 50% of the total landed value. Its management system shares many main characteristics with other managed fisheries in the developed world. A total allowable catch (TAC) system was introduced gradually for cod and the other main species during the 1980s, to protect fish stocks and stabilize landings (Hannesson 2013; Standal, Sønvisen, and Asche 2016).2 As shown in Figure 1,3 since 1995 the landing value in real terms for groundfish has remained relatively stable at about NOK 6 billion per year, although with substantial annual variation. During the same period, the number of landing plants has been reduced by nearly 50%, from 370 to 205. This is not surprising given that the real income level in Norway is increasing (SSB 2017), and with a given value of landings, productivity must increase if the plant workers are to continue earning an in-come competitive with alternative occupations. Over time, the number of fishers and the number of vessels have also been reduced for similar reasons, and possibly more so due to the gradual movement toward individual fishing quotas (IFQs) (Standal and Hersoug 2015; Standal, Sønvisen, and Asche 2016; Hannesson 2017).
Number of Landing Plants and Landing Value for Groundfish in Norway
The fishery management system in Norway maintains an owner-operated coastal fleet (Standal and Hersoug 2015), making it illegal for landing plants to own fishing vessels.4 Moreover, the fisheries regulations in the groundfish sector favor coastal vessels. For instance, to maintain a fleet structure dominated by coastal vessels and to help prevent geographical concentration in landings, about 70% of the cod quota is allocated to coastal vessels in a formula known as the trawl-ladder, where the coastal fleet receives a higher share of the TAC in years with a lower TAC (Guttormsen and Roll 2011). Fishers are also protected from oligopsonistic buyer behavior by sales organizations with a monopoly on first-hand sale of fish.5 Any buyer must register with the sales organizations that receive payment on behalf of the fishers, allowing them in turn to enforce a minimum price.6
The data used in this paper have been collected and provided by the Norwegian Fisher-men’s Sales Organization (Norges Råfisklag; NFSO),7 the largest of the six sales organizations, whose area extends from the northern-most region of Finnmark to Nord-Møre (Appendix Figure A1). The NFSO holds the responsibility for all landings of fish and seafood in the northern half of Norway, with the exception of pelagic fish. The raw data contain trip-level landing records for all commercial species transacted under the auspice of the NFSO between 2002 and 2015.
The dataset was aggregated to the level of landing plant, on a yearly basis. The landings cover more than 10 target species, some of which are managed with IFQs, some are regulated or restricted open access, some are open access, and a number are less important bycatch species. The main species landed are shown in Figure 2. Cod is by far the most important one by quantity, and even more so when considering value, since it is the high-est-priced groundfish species. Haddock and saithe are also of significant importance and together with cod account for over 80% of the landed value.
Average Yearly Quantity Landed, by Species
The coastal fleet is highly dependent on the migration patterns associated with spawning aggregations for the main species, of which the most well known is that for cod in Lofoten during winter and early spring months (Hannesson, Salvanes, and Squires 2010; Kvamsdal 2016). This gives rise to high seasonality in the fishery, as shown for cod in Figure 3, which creates challenges for capacity utilization at a landing plant. While this issue has not in itself received much attention, with Matulich, Mittelhammer, and Reberte (1996) as a notable exception, it is largely a reflection of the overcapacity in the fleet, leading to derby fisheries and short harvesting seasons. A Herfindahl-Hirschman index (HHI) was constructed to measure concentration of landings over the year for the various plants. This is graphed, by landed quantity, in Figure 4. When seasonality is low and landings are spread out equally over all months, the HHI will have a lower bound at 1/12, and it will be equal to 1 if all fish are landed in the same month. This is the case for some landing plants (Figure 4). Similarly, an additional HHI was computed to measure each plant’s landings concentration over fish species, to account for the diversity of landed species. In contrast to the HHI for season, the impact of the HHI for species is not so straightforward, as diversification can be a tool to reduce risk, while specialization can reduce cost.
Seasonal Landings Pattern for Cod, 2002–2015; Landings Normalized to Average Monthly Landings across Years
Concentration of Landings over the Year, by Plant Size
Several vessel and gear types are used in the Norwegian groundfish fleet. Figure 5 shows the annual average landings by gear type, with the number of vessels using the associated gear at the top of each bar. Nets are the gear catching most fish, followed by Danish seine, trawl, and hook and line. Nets and Danish seine are the gear types used by most of the larger coastal vessels, and their impor-tance for the landings reflects the importance of these vessels in the fishery. Hook and line is the preferred gear type for the smallest vessels, the most abundant in terms of number of vessels. Hence, hook and line are the gear type used by most vessels, although they are less important in terms of total landings due to the limited landings per vessel. The trawlers and long-liners land significant quantities despite the relatively low number of vessels, reflecting these vessels’ fishing power.
Average Yearly Landings by Gear Category, 2002–2015; Average Number of Vessels Are above Bars, and Average Vessel Length Is per Category
An individual landing plant is defined as a plant receiving fish at a specific geographic location that may or may not conduct additional processing of the fish. In cases where a plant was observed to not receive landings for one year or longer, its reappearance in the dataset was defined as a new landing plant. The motivation for this decision is based on the fact that landing plants may file for bankruptcy and possibly experience a change in ownership and go through significant management and strategy changes or renovation, in which case they can be regarded as a new plant. Communication with the NFSO indicates that most plants that reopen do so under new ownership.
There are other attributes that may play meaningful roles for a landing plant’s continued operation, and that accordingly have been included in the analysis. A measure of a landing plant’s attractiveness is the extent to which it is able to obtain landings from nonlocal vessels. Conversely, a high proportion of local vessels landing at the plant can be viewed as a measure of vessel loyalty. The landed quantity influences a landing plant’s size and its opportunity to exploit economies of scale. The number of buyers, forming links in the supply chain, can be an indicator of a plant’s attractiveness and of the competition for its output. Moreover, the number of additional landing plants within the same municipality is a measure of a possible industry cluster in that region. In addition to providing external economies of scale, industry clusters may be attractive to fishers because they increase competition for their landings. Horizontal integration may also play a decisive role in a plant’s survival due to the parent company’s strategic influence over the different landing plants owned.8
Descriptive statistics for all variables used are presented in Appendix Table A1.
The data are left truncated and right censored (LTRC). To address these issues, all observations of landing plants that appeared only in 2002 were removed from the sample due to the fact that the available dataset does not provide any information of when those plants first entered the market. Similarly, data for 2015 were used to find out if the landing plants active in 2014 remained active or exited the market after 2014.9 Hence, the estimation sample spans the period from 2002 to 2014.
3. Methods
First, a pooled logit model is used to estimate the probability that a landing plant is exiting the industry. This model is then extended to a conditional logit model (McFadden 1974), where the response can be interpreted as the landing plants’ decision to continue in business or to close down or exit. The conditional logit models introduce two sets of fixed effects. The TAC, which is set on a year-by-year basis, is likely to influence the landings of fish at all plants. In addition, some areas may in-crease their probability of remaining active, either because they are closer to the fishing grounds or due to better logistics. To control for all time-varying and location-varying factors that would influence a landing plant’s activity, the model was run with both year fixed effects and with year-region fixed effects. The definition of region was aligned with the areas under the auspices of NFSO control. To correct for the tendency of continuous variables toward right-skewedness, a log-transformation was applied.
Duration analysis (Wooldridge 2010) is used to investigate factors that influence the length of any spell, and in this case how long a landing plant remains active. The survival function is defined as
where N measures the number of consecutive years a landing plant remains active. The sur-vival function is non-parametrically estimated using a Kaplan-Meier filter. The hazard function, or the conditional failure rate, measuring the probability that a landing plant exits the industry after time t given that it has been in business up to time t, can be written as
The same attributes as in the (conditional) logit analysis are used as potential factors that influence the survival of landing plants. The hazard rates are estimated using the partial likelihood method proposed by Cox (1972): where h0 is the baseline hazard function at time t, while x represents the vector of time-varying covariates. The Cox function allows the β parameters to be estimated without estimating the baseline hazard. The model is estimated with and without region fixed effects. Moreover, reentering plants are also accounted for in the analysis using multiple-spell dummies.10
4. Empirical Results
The estimation results for the pooled logit (Model 1) and the conditional logit (Model 2 and Model 3) models are reported in Table 1.11 Across models, there is strong evidence in support of landings, landing plant clusters, and seasonality playing a role in the survival of landing plants.
Results from the Pooled Logit (Model 1) and the Conditional Logit (Model 2 and Model 3) Models with Robust Standard Errors and Bootstrapped Standard Errors, Respectively
The coefficients on the landings variable are consistently negative and statistically significant, indicating that larger plants have a lower probability of exit. Robust standard errors are provided for the pooled logistic model (Model 1), and bootstrapped standard errors (Wooldridge 2010; Cameron and Miller 2015) are reported for Models 2 and 3.12 Models 1 and 3 also indicate that the number of additional landing plants within the same municipality reduce the probability of exit, suggesting that there may be an industry cluster effect.13 Plants that consistently receive landings from smaller vessels (under 15 m) have a lower probability of exit. This is an indication that at least for some plants, the coastal fleet is important and that for these plants the smaller vessels support the community through the continued operation of the plant. It is interesting to note that being able to attract larger, more mobile vessels, or vessels from a variety of municipalities, does not influence the probability of exit. Receiving landings over a shorter period of time, not surprisingly, increases the chance of an exit. However, there is no specialization/diversification effect. Finally, the number of buyers being served by the plant and the existence of horizontal integration do not present an association with the probability of exit.
Estimation results for the four duration models are reported in Table 2. The reported parameters associated with each variable represent hazard ratios, calculated by exponentiating the estimated coefficients. Hence, a hazard ratio higher than one increases the probability of a failure, while a hazard ratio smaller than one reduces it. The significance level reported is therefore for the null hypothesis that the reported hazard ratio equals one. The downward-sloping survival functions in Appendix Figure A2 suggest negative duration dependency.14 That is, the longer the landing plant has been active, the lower the probability of it failing. Moreover, it is evident based on Appendix Figure A2 that only a small fraction of the firms fail after one year of operation. For both Models 4 and 6 the 10year survival rate is approximately 50%.
Results from the Survival Models on the Right-Censored Data
The results from the duration models indicate that all attributes that were originally found significant in the logit models remain significant, and in the expected directions. For example, the hazard ratio for the number of additional landing plants in the same municipality is below one, implying that a lower number of landing plants in a municipality leads to an increased chance of failure. Hence, the empirical results appear robust relative to estimation procedures.
5. Conclusions
Most of the attention in fisheries economics is given to what happens on the water. However, this is just the first step in the seafood supply chain. As demonstrated by Homans and Wilen (2005), which supply chain is served can have substantial impact on fishers’ income. More important for fisheries policy is that where the fish are landed strongly influences the impact of the industry in terms of jobs and economic activity in coastal communities. While it is well known that the number of fishers and vessels are being rapidly reduced in many countries, little attention has been given to the reduction in the number of landing spots and thereby their impact on coastal communities.
Norway is an example of such a country where the number of fishers as well as the number of landing plants are being reduced. Using data for the groundfish fisheries, this paper investigates if there are specific factors influencing which plants exit the industry. The results indicate that the size of a landing plant, as measured by total quantity landed, and landings from a fleet of smaller vessels reduce a landing plant’s probability of failure. In addition, stronger seasonality in landings increases the odds of exit by reducing capacity utilization at plants. Perhaps the most interesting result is that plants located in municipalities with additional landing spots have a lower probability of exit. This illustrates that industry clusters may be important. However, community size as measured by population count does not influence survival probability, and there is accordingly no indication that the cluster benefits much from activities that are not fishery related. Hence, there appears to be no reason for the industry cluster to move to a larger city. It is also interesting that being able to attract vessels from other regions, or more mobile vessels does not influence the probability of a landing plant’s survival.
Overall, the results suggest a bleak future for many vulnerable coastal communities. With most fisheries being fully or overexploited (FAO 2016), the only way some landing plants can increase landings is through the disappearance of others. Extended fishing seasons will also contribute in this respect. This is, however, not an entirely bad outcome, as it provides job security for those who remain employed, not unlike what Abbott, Garber-Younts, and Wilen (2010) report following a fisheries rationalization program that extended the fishing season. That the fisheries cluster seems to be independent of other in-dustries also implies a way forward for fisheries-dependent coastal communities that want to thrive. This requires policies that support the formation of clusters, maintain a coastal fleet, and support management changes that extend the fishing season.
Acknowledgments
The authors thank two referees for helpful comments and acknowledge financial support from the Norwegian Research Council (CT #233813) and the Norwegian Seafood Research Fund (CT #901266).
Footnotes
↵1 Coastal communities that were once built around fishing-related activities are now often perceived to experience a loss of “values” when fishing no longer serves as their principal occupation (Blythe 2015; Tull, Metcalf, and Gray 2016; Thompson, Johnson, and Hanes 2016).
↵2 The TAC was first introduced for trawlers in 1981 and was extended to all vessels by 1989.
↵3 The figure combines data provided by Fiskeridirektoratet (Fisheries Directorate of Norway) and data from NOFIMA (a Norwegian research institute). Fiskeridirektoratet provides total landings of groundfish in Norway (publicly available at https://www.fiskeridir.no/Yrkesfiske/Statistikkyrkesfiske/Fangst-og-kvoter/Norges-fiskerier, accessed November 4, 2016). The annual numbers of landing spots in Norway were provided by NOFIMA and are based on a survey they conduct on behalf of the Norwegian government; data access was arranged through personal communication with Bent Dreyer.
↵4 There exist a few exceptions for the trawler group. Two companies are allowed to own trawlers with the objective of ensuring that the plants receive sufficient quantities of fish.
↵5 There are six sales organizations: one that handles all pelagic species, and five regional organizations that handle all other fish species.
↵6 While the sales organizations have monopsony power, there is no evidence that this is exploited beyond protecting fishers from potentially oligopsonistic buyers (Asche, Chen, and Smith 2015; Pettersen, and Myrland 2016).
↵7 Researchers may make a formal request to the NFSO for specific data. The exact dataset used in this analysis is available from the authors for purposes of replication.
↵8 Vertical integration may affect the survival of landing plants, since the decision of closing a particular plant is likely to depend on a larger multiplant operating plan. However, data in this regard is not available and, consequently, could not be incorporated into the analysis.
↵9 About two-thirds of the observations relate to plants that had been in the study since at least the first year of obser-vation, creating a left-truncation challenge. A common approach used to address left-truncation is to remove those observations. This is not feasible when such a large portion of the data would have to be deleted. Alternatively, the non-parametric tree estimation method of Fu and Simonoff (2017) could be used if information on the age of each plant at entry in the study were available. As a robustness check, survival models with all these observations removed were run with an obvious loss in power. All variables previously significant remained relevant, to a large degree. These results are provided in Appendix Table A2.
↵10 The data do not provide the cause for a reentry, and therefore a clear differentiation between what entails a new entry versus what qualifies as a reentry is not possible.
↵11 The number of observations used in the analysis is adjusted when fixed effects are included, due to multiple positive or negative outcomes being encountered within groups.
↵12 The bootstrapping was conducted with 400 repetitions and seed = 10101. In the tables, all corrected standard errors are referred to as robust standard errors.
↵13 To our knowledge, there exist no studies investigating industry clusters in fisheries. Tveteras (2002) and Asche, Roll, and Tveteras (2016) provide evidence of industry clusters in the Norwegian aquaculture industry.
↵14 Only duration Models 4 and 6 are presented, for readability. When fixed effects are considered, the plots become more cumbersome, as there is a function associated with each fixed effect, with no added value to the interpretation.
