Economic Impacts of Marine Reserves: The Importance of Spatial Behavior Martin D. Smith Assistant Professor of Environmental Economics Nicholas School of the Environment and Earth Sciences Duke University Box 90328 Durham, NC 27708 (919) 613-8028 marsmith@duke.edu and James E. Wilen Professor Department of Agricultural and Resource Economics University of California, Davis One Shields Avenue Davis, CA 95616 (530) 752-6093 wilen@primal.ucdavis.edu JEL Codes: Q22 Short Abstract Marine biologists have shown virtually unqualified support for managing fisheries with marine reserves, signifying a new resource management paradigm that recognizes the importance of spatial processes in exploited systems. Most modeling of reserves employs simplifying assumptions about the behavior of fishermen in response to spatial closures. We show that realistic depiction of the behavior of fishermen matters dramatically to the conclusions about reserves. We develop, estimate, and calibrate an integrated bioeconomic model of the sea urchin fishery in northern California and use it to simulate reserve policies, but with behavioral response via reallocation of effort over space. Our behavioral models show how economic incentives determine both participation and location choices of fishermen. We compare our results with biological modeling that presumes effort is uniformly distributed and unresponsive to economic incentives. The modeling shows that the optimistic conclusions about reserves may be an artifact of simplifying assumptions that ignore economic behavior. February 5, 2002 Economic Impacts of Marine Reserves: The Importance of Spatial Behavior Long Abstract Resource scientists have recently shown virtually unqualified support for managing fisheries with marine reserves, signifying a new resource management paradigm that recognizes the importance of spatial processes in both untouched and exploited systems. Biologists promoting reserves have based such support on simplifying assumptions about harvester behavior. This paper shows that these naïve assumptions about the spatial distribution of fishing effort before and after reserve creation severely bias predicted outcomes, generally overstating the beneficial effects of reserves. The paper presents a fully integrated, spatial bioeconomic model of the northern California red sea urchin fishery. The model is the first attempt to marry a spatially explicit metapopulation model of a fishery with an empirical economic model of harvester behavior. The biological model is calibrated with parameters representing best available knowledge of natality, growth, mortality, and oceanographic dispersal mechanisms. The model of spatial behavior is estimated using a large panel data set of urchin harvester decisions, which are recorded in logbooks and on landings tickets. The economic model tracks repeated decisions that take place on different time and spatial scales. Harvesters make daily decisions about whether to fish and where to fish, conditioned on home ports. On longer time scales, harvesters select home ports, choose between northern and southern California regions, and decide whether to drop out of the fishery. Combining a Random Utility Model with Seemingly Unrelated Regressions, the model predicts the aggregate supply response of fishing effort and the 1 spatial allocation of effort as a function of relative economic payoffs. The economic model is integrated with the biological model to endogenize effort, harvest, biomass levels, and reproductive performance of the population. The integrated model is used to explore the efficacy of marine reserve formation under realistic assumptions about the determinants of total effort and its spatial distribution. Contrary to the dominant biological modeling conclusions based on naïve assumptions, realistic behavioral assumptions lead mostly to pessimistic conclusions about the potential of marine reserves. The results ultimately cast doubt on many of the arguments made by those who advocate 2 Economic Impacts of Marine Reserves: the Importance of Spatial Behavior I. Introduction There is an important paradigm shift underway in the marine policy arena that is likely to change the way in which coastal resources are managed in the future. The shift is toward the use of spatial zoning measures that will effectively carve the ocean up into a system of mixed areas of regulated exploitation and areas protected with marine reserves.1 The underpinnings of this new vision are based on a mix of emerging science and political interest in marine protected areas. On the science side, there is growing consensus among marine ecologists, biologists, and many fisheries managers that conventional season length and gear restriction management methods have failed and are bound to fail in the future, and that a new approach is therefore needed. On the political side, there is a growing view among NGOs and influential environmental lobbying groups that long term biodiversity and conservation goals would be best served by a network of protected areas similar to our terrestrial park systems. Up to this point, the case for this new spatial zoning view has mainly been promoted by biologists, with little input solicited from economists and other policy analysts. But as real proposals emerge for specific systems of protected areas, there will be increasing calls for economic analysis of impacts of actual policy options. Economists are only beginning to think about how space might be introduced into conventional models of renewable resources and how spatially differentiated policies might compare with second best undifferentiated policies. In this paper, we present a comprehensive empirical investigation of the implications of marine reserves, as they might be applied in a case study of the northern California red sea urchin fishery. We focus particularly on modeling spatial behavior of the harvesters, and we assess the degree to which accounting for economically-driven behavior matters to the conclusions coming out of recent biological investigations of marine reserves. We find, a bit to our surprise, that the many and varied simplifying modeling assumptions made by biologists to handle harvester behavior do not actually “cancel out” in the final analysis. Instead, virtually every simplification made in biological modeling about economic behavior biases the case in favor of the use of reserves as a fisheries management instrument. Our results thus call into question whether the optimism displayed for reserves as a fisheries management tool is warranted. In the next section we review what is known about reserves, from both the biological and economic literature. We then describe the sea urchin case study fishery, focusing particularly on its spatial character and its short- and long-term dynamics. In the fourth section we discuss and estimate some models of spatial behavior and then use those, in the fifth section, to simulate the impacts of marine reserves with an integrated 1 The emerging literature on marine reserves uses a variety of terms that are sometimes synonymous and sometimes not. We will use marine protected areas, marine reserves, and no-take zones as interchangeably referring to areas in which no exploitation is permitted, including especially commercial and recreational fisheries exploitation. 3 bioeconomic model. In the final section we draw some broad conclusions for both research and policy analysis of marine protected areas. II. Related Biological and Economic Work The notion of using closed areas as a marine management tool emerged a little over a decade ago among marine ecologists and conservation biologists. The first proposals were modest, calling for small areas off coastal research institutes in which ecologists could study unexploited systems in order to gauge the ecological impacts of exploitation. By the early 1990s, the idea was beginning to morph into a grander vision that called for significant areas to be set aside, often on the order of 20-30% of the coastline. The transformation in the scale of the proposals coincided with several important papers on fisheries management, most of which concluded: a) that the world’s fisheries were in a state of crisis; and b) that conventional methods were to blame and that a new approach to management was needed. 2 Early modeling papers thus focused on the impact that permanent closures might have on exploited fisheries. The first modeling work on marine reserves is by Polacheck, who used a Beverton/Holt yield per recruit model to examine how a purposeful closure might affect an exploited fishery. Beverton/Holt developed the yield per recruit model in the 1940s to describe fisheries in which the relationship between the spawning stock and the resultant recruitment of juveniles is highly uncertain or unknown. Yield per recruit models assume larval recruitment is constant and exogenous and then the equilibrium is expressed in terms of two policy variables of interest, spawning stock biomass per recruit (SSB/R) and yield per recruit (Y/R). In comparing management policies, options that involve more Y/R are generally judged as desirable in terms of fishery objectives, as are options that increase SSB/R.3 Choices that increase one and decrease the other are problematic, however. In Polacheck’s first simulations of reserves, SSB/R always increases over the whole system with a reserve, but Y/R generally decreases in most simulations.4 Polacheck’s and other similar yield per recruit modeling exercises were an important first step in understanding how marine reserves might work as a management tool. Most of these “first generation” models assume: a) a homogeneous area that is subdivided by a reserve into two areas; b) a constant amount of effort redistributes to the remaining open area, thus increasing fishing mortality in the open area; c) simple 2 See for example the oft-cited paper by Ludwig and Walters (1993), which argues that reductionist approaches cannot overcome the natural variation and irreducible uncertainty inherent in natural systems in order to guide us toward anything like sustainable yield. 3 With a fixed or unknown stock/recruitment relationship, it is obvious why having more fisheries yield per recruit is desirable. A caveat is that the B/H models are equilibrium models and, from a dynamic perspective it might not be worth the sacrifice to restore a fishery to the point that squeezed the last equilibrium levels of yield out of the system. Having more SSB/R is judged desirable because if there is any positive relationship between spawning stocks and recruitment, a higher SSB/R would increase recruitment and yield. Moreover, many fisheries scientists point to the “safety” provided by higher levels of SSB in the face of perturbations and environmental shocks. 4 Polacheck concludes “in some cases the yield per recruit with a refuge can exceed the yield per recruit without one, but the net increases are usually small”. He also concludes that “a closed area can lead to substantial increases in spawnng stock biomass. . .and, as such, could be a viable short term management option to reduce overall fishing mortality on an overexploited stock.” 4 dispersal assumptions such as proportional diffusion of adults; d) equilibrium analyses that ignore transitional effects; e) no explicit relationship between spawning stock biomass and recruitment of larvae in the system. Despite simplifying assumptions embedded in these modeling efforts, compilation of empirical data on closed areas around the world has confirmed most of the modeling predictions, at least of effects within the closed areas. A recent metapopulation analysis of 89 papers with empirical data on reserves confirms the virtually universal increase in spawning stock biomass produced within reserves that was first predicted by Polacheck and by other similar yield per recruit models.5 One does not need a voluminous amount of empirical work to believe that removing fisheries exploitation from an ecosystem will increase biomass, broaden the age distribution by providing protection for larger and more fecund fish, and perhaps increase biodiversity and alter species composition within the protected area. What is less certain, however, is whether the increase in spawning stock biomass and its composition within a reserve can provide a net increase to the fishery outside the reserve.6 The first generation modeling work identifies some mechanisms that can be important to this question, namely the mobility of adults and the level of pre- and post-reserve exploitation in the fishery. The first generation papers show that reserves are most likely to increase yield in the open area when adult mobility is not too low or too high. The intuition is that, for a given size reserve, adults must not move so widely that they are not afforded protection by the reserve, but some adults (and juveniles) must move widely enough so as to increase the fishable abundance available to the fishery. The second generation of papers generalized the yield per recruit models by closing the relationship between spawning biomass and recruits to the fishery. This is an important addition to understanding, because protected areas may also produce eggs and larvae that are subsequently distributed to exploitable populations in the remaining open areas. In particular, a small number of papers from the late 1990s assume that adults produce larvae, which disperse in some manner and then recruit into both the fishery and the reproductive population. One of the most significant second generation papers is the paper by Holland and Brazee, two economists. The Holland/Brazee paper is sophisticated from a biological modeling point of view, incorporating density dependent stock/recruitment relationships in both the reserve and open area, and uniform larval dispersal mechanisms in a detailed age-structured population model. In addition, their model incorporates economic variables and is fully dynamic so that the present value of transition paths can be examined. The Holland/Brazee analysis confirms the Polacheck results that spawning stock biomass will increase with reserves. They also find that whether this spills over sufficiently to produce a net economic gain depends upon the discount rate and the pre-reserve exploitation level. At high discount rates it never pays to endure the sacrifice necessary to rebuild sustainable harvests to higher levels. In contrast, at high pre-reserve exploitation rates, since the overall harvest is driven so far 5 Halpern, B. (2002) Some of the fisheries literature does not seem to understand that it is a net increase and not just a gross increase in yield that is needed in the remaining open areas. That is, from the fishing industry perspective, it is not enough that a reserve increases harvest outside the reserve, but rather that the increase be large enough to compensate for the area removed from fishing. As the reserve size gets larger, of course, this hurdle rises. 6 5 below MSY, it is more likely to be worth an investment in rebuilding the overall biomass by using a reserve.7 One important simplifying assumption made by Holland/Brazee and virtually all of the analysis of reserves that has been done by biologists is that effort is fixed both before and after reserve formation. Under a closure, it is presumed that the effort that is transferred to the remaining open area increases fishing mortality there, and in the most sophisticated analysis there are further ramifications on harvest, biomass, and reproduction as a result. In this paper we consider how incorporating more realistic depictions of harvester behavior affects the potential implications of marine spatial closures. A priori, economically motivated harvester behavior ought to matter in several ways. First, since the pre-reserve status quo is important to the net effect of a policy change, it should be of interest to know how economic variables condition the initial circumstances in a spatial bioeconomic system. Second, since a spatial closure will affect the subsequent spatial distribution of relative economic returns, we would expect that effort redistribution would have complicated spatial and intertemporal effects, both in the short run and in the long run. Third, since most real spatial systems embody complicated heterogeneities that affect profit differentials over space, we would expect to be able to capture some of these as profit differentials. For all of these reasons, simplified assumptions about effort distribution and its determinants are likely to confound and possibly mislead policy makers considering these new forms of marine policy instruments. III. Case Study: the Northern California Sea Urchin Fishery The Northern California sea urchin fishery is an ideal case study with which to examine empirically marine reserves and the dynamics of spatial behavior of harvesters. Sea urchin are found along the Pacific Coast in rocky inter-tidal areas kelp forest areas from Southern California to Alaska. Urchin have hard spiny shells and they are harvested for the gonads (roe) found inside the shells. Urchin are harvested by divers who make day trips to fishing grounds on vessels built especially to travel fast to the dive location. Once on a site, a diver begins harvesting while a tender aboard the vessel watches the scuba gear. Harvesters use rakes to remove urchins from the rocky bottom, filling mesh toat bags which are then offloaded onto the vessel by the tender. Weather conditions are a critical determinant of when and where divers choose to dive, particularly in Northern California where a trip is typically made only 14% of available open days. At the end of a fishing day, harvesters deliver the whole urchins to processors. At the processing plant, workers split the shells, scoop out the gonads, and then wash and bath the gonads in alum solutions to firm the roe. Roe skeins are then carefully packed in special wooden trays holding 250g. and the product is shipped overnight to Japan. In 7 At the same time, what this shows is that it can be a sound investment to rebuild an overexploited stock by reducing effort. Closing a fraction of a fishery’s area is one way to reduce overall fishing mortality (as Polacheck concluded), but another way is to simply crank down conventional methods of effort and fishing mortality control. Hastings and Botsford have established that area controls are equivalent to conventional effort controls under certain reasonable circumstances. Some fisheries observers point out, however, that closed areas may be easier to enforce than conventional effort control measures such as mesh size, days at sea, etc. 6 Japan, the roe is sold mainly at the Tokyo Central Wholesale Market, in competition with roe from Japan and other North American fisheries. The roe is relatively high valued, with current Tokyo Wholesale prices in the range of $25 per pound, translating into a markdown to divers of approximately one dollar per pound of whole urchin.8 A typical day trip brings in 750 pounds of urchin, grossing $750 per trip for the diver/tender team. Urchin have been harvested in Southern California since the early 1970s and in Northern California since 1988.9 They are regulated with a combination of closed seasons, minimum size regulations, and a limited entry program. Current regulations require that Northern California urchin be at least 3.5 inches in diameter, a size reached at an age of approximately – years. A full closure is in effect throughout all of July, and one week closures are operative from May to September. Within each open week, threeday per week openings are in force in June and August, and four-day openings prevail in April and October. In total, the number of potential open days per year is approximately – in Northern California. The limited entry program was introduced into the Californiawide fishery in 1989, grandfathering all existing permit holders (roughly 850) into the fishery. The program has been tightened up over the intervening years in order to steer participation to a long term goal of 300 divers. Each diver must now land 300 pounds per trip for a minimum of 20 trips during either the current or preceding year in order to be granted a license.10 Licenses must be fished by the holder, and licenses may not be leased or sold. Biological characteristics of urchin mimic those identified by biological modelers as holding the most promise for successful use of reserves. First, urchin are “patchy” in that they are found in several distinct and discrete areas in Northern California where substrate and habitat characteristics are suitable for the species. Second, adult movement within each patch is relatively low, an average of 7-15 cm. per day. Third, closed areas promise protection of a substantial amount of productive spawning biomass. Urchin reach sexual maturity approximately at age six, and egg production increases with age at an exponential rate. Fourth, larvae are redistributed considerable distances from spawning areas by currents, winds, and sea surface changes. In addition to favorable biological characteristics, the urchin fishery is also an ideal case study for examining spatial behavior of harvesters. Most importantly, there are multiple dimensions to the spatial choices divers face. Among the limited set of licensed divers allowed to participate, divers each make seasonal decisions about whether to fish in Southern California or Northern California. These large scale spatial decisions are made infrequently, perhaps averaging once or twice per diver over the whole period we examine. Additionally, for those divers choosing Northern California, each diver must make a port choice decision, essentially a decision about where to deliver (and where to live) during the fishing season. This is an intermediate scale spatial decision, and it 8 Note that the recovery of roe from whole urchin is about 10%. Hence ten pounds of whole urchin are needed to generate a pound of roe, mostly explaining the markdown. 9 Ironically, urchin were considered pests by abalone divers in the 1960s because they competed for the lucrative abalone for habitat and food. Abalone divers often poured quicklime on urchin in order to remove them and open up habitat for the commercially valuable abalone. In the 1970s, a market for California urchin developed as the Japanese stocks were overharvested. The fishery was harvested mainly in Southern California by ex-abalone divers after the abalone fishery collapsed. The continued growth in the Japanese market led to the Northern California fishery being opened up in the late 1980s. 10 Provisions are also made to allow one new entrant for each ten licenses retired. 7 determines the reach and the nature of patches available for day trips taken from each respective port. Finally, each diver leaving from a particular port on a daily basis chooses which patch to visit from among those feasible within a day trip’s distance. These kinds of decisions obviously cover the smallest geographic scale and over a most frequent time scale. The data set we use to examine spatial harvester behavior in the Northern California sea urchin fishery is unusually rich. It consists of 57,000 individual dives made by up to 358 divers over the period between 1988 and 1997, recorded in mandatory logbooks and landings ticket records. Each landings ticket record records port of landing, processor code, quantity landed, price paid, and diver code. Each logbook record reports dive location (latitude), dive duration, average depth, and divers per vessel. We also collected daily weather data on wind speed, wave height, and wave period from weather buoy records, averaged it over the period preceding a typical trip decision, and keyed it to potential dive trip decision. The fact that each trip is a day trip is also convenient since we can model decisions as repeated nested discrete choices.11 The models we report here estimate parameters of a repeated choice structure for several hundred divers making decisions over a total of over 400,000 choice occasions.12 During this period, the urchin fishery was harvested from a virgin fishery to the present level approaching a steady state at average profit levels. Prices vary over the period as a result of exchange rate changes, quality changes, and landings variability. Weather conditions vary on a daily basis and with some inter-annual variation as well. As a first step in our analysis of the urchin fishery, we examined the nature of spatial behavior by divers taking day trips in Northern California between 1988 and 1997. We first identified six ports that handled urchin and then keyed latitude-based diving locations to those ports. Table 1 shows some of the raw data and it reveals that diver activity off each port is concentrated in nearby locations, some of which overlap for several ports. For example, most of the urchin landed in the southern- and northern-most ports of Half Moon Bay and Crescent City are taken in patches located directly offshore of the port location latitude. In contrast, urchin landed at Fort Bragg are harvested mostly from four nearby patches, two of which are also major sources for Albion, an adjacent port. Ports of Bodega and Point Arena land urchin from several nearby locations, although with relatively little overlap. What these show is that the spatial choice set as revealed by actual behavior varies with port location, with landings fanning out in a Laplacian fashion that declines exponentially from distance off the port. Some of these spatial landings patterns also reflect a pushing out of the Ricardian margins, in that early exploitation was shallow and near ports whereas later exploitation has been deeper and at more distant patches. Using these effort data and knowledge of breaks in suitable habitat types, we partitioned the entire data set into one characterized by eleven patches along the coast of Northern California. Figures 1 and 2 are a histogram and map over the whole period with 11 A repeated decision structure is more tractable than trying to model situations with a fixed end point. In addition, fisheries in which multi-day and multi-area trips are made must contend with the complications introduced by decisions that are essentially searching decisions made along the journey to a targeted destination. 12 Total choice occasions are days for which the urchin fishery is open to harvesting. Because weather conditions so frequently keep fishermen ashore, the actual number of trips is only a fraction of total choice occasions, about 14% of open days. 8 diving activity characterized by “patch” number.13 This shows the large concentrations of effort located near the four ports with highest landings, together with breaks between patches associated with substrate and habitat variation. Tables 2 and 3 show the extent of spatial mobility across patches in Northern California, and between Southern and Northern California on an annual basis over the whole sample period. Several interesting facts can be gleaned from these tables. One can see the open access implications of opening up the previously unexploited Northern California fishery starting in 1988. From zero activity in 1987, harvesters from the pool of divers in the Southern California fishery and elsewhere were drawn rapidly into the new fishery, with over 100 new divers participating in 1988. The Northern California part of the fishery peaked in 1992 with 358 total divers, of which 108 fished in both regions during the year. During the peak period, some divers fished in as many as 6-8 Northern California patches during the year, although the majority concentrated on one or two patches from a single port. As the Northern California fishery matured and as the unexploited stocks were drawn down and profitability reduced, fishermen exited, some returning to the Southern California fishery and other dropping out entirely in response to the stringent limited entry requirements. Total California-wide numbers dropped from 826 in 1994 to 440 in 1998, whereas in Northern California, numbers dropped from 358 to about 125 in recent years. Table 4 shows intra-port mobility over the whole sample period. The mobility rate can be computed by noting that there are 57,550 dives reported in Northern California ports for which the diver previously landed in a Northern California port. Of these, 2,796 were occasions in which the diver landed in another Northern California port, or about 5% of the occasions. Correspondingly, there were a total of 167,042 total dives made in both Southern and Northern California in the sample period, including 1682 switches across regions, or about 1%. In the last couple of years, there has been a slight reduction in average mobility across Northern California ports, and a marked reduction in large scale mobility between Southern and Northern California regions compared with the peak period. The Northern California fishery appears to have settled into a pattern approximating a bioeconomic equilibrium, with harvesters fishing more intensively to maintain a living under reduced abundance conditions. The average number of trips per year has increased from about 26 per year in 1992/93 to a current level of about 34 trips per year. Average depth per dive (mean 35.45 feet over all years/locations) has similarly increased at a rate of about one foot per year as the extensive margins have been pushed out from all port locations. IV. Modeling Spatial Behavior In this section, we report results using the logbook/landings ticket-based data set on divers’ daily spatial behavior to estimate several models of individual choice. It is convenient to begin with the hypothesis that individual divers make daily choices that maximize some random utility function: Uijt = vijt + εijt 13 We do not show the eleventh and southern-most area, which is off the Farallon Island off San Francisco Bay. 9 = f(Xit, Zi1t, Zi2t, ... , ZiMt; θ) + εijt , (1) where Xit includes harvester-specific and time-specific characteristics that are constant across choices, Zimt includes choice-specific characteristics such as travel costs and resource abundance, θ is a parameter vector, and εijt is a random component that is unobservable to the analyst. The random utility model posits that, given M possible dive locations, diver i in period t will choose location k if the utility of choice k is higher than that of the other M-1 choices as well as the choice of not to dive in period t . There are numerous discrete formulations that capture the essence of spatial decision-making and are consistent with the random utility formulation. General approaches fall into the following categories: Multinomial (and Conditional) Logit (MNL), Discrete Choice Dynamic Programming (DCDP), Random Parameters Logit (RPL), Multinomial Probit (MNP) and Nested Logit. We use a Repeated Nested Logit (RNL) approach in this analysis for various reasons.14 First, it can be argued that the structure of the RNL mimics the decision problem faced by fishermen in this industry reasonably well. In particular, fishermen actually do make repeated decisions as if there is no time horizon and without the need to make complicated dynamic decisions. Essentially what changes from day to day are variables that determine short term expected revenues and variables that determine weather-related risks associated with participation. Second, the RNL is relatively easy to estimate with the large size data set we have. Third, we need an estimation scheme with a closed form solution in order to embed the estimated behavioral choice model into the bioeconomic simulation model discussed in the next section. The RNL also has well-known properties that are desirable in general, including the flexibility to admit different variances at different decision nodes.15 We follow McFadden (1978) by assuming that the εijt is independently and identically distributed Generalized Extreme Value, and that the utility is linear in individual- and choice-specific variables so that the following model characterizes individual choices: z jt ' γ exp + x t ' β + (1 − σ)I (1 − σ) Pr(Go to j ) = 10 z kt ' γ z kt ' γ + x'β + (1 − σ)I + exp exp ∑ k =0 (1 − σ) (1 − σ) (2) and 14 On the surface, the choice structure is the same as the RNL proposed by Morey et al. (1993), which deals with repeated participation and site choice in recreation demand. 15 The drawbacks of RNL are equally well known and include inflexible incorporation of the possibility of heterogeneity across individuals, and assumptions of independence across time. 10 10 Pr (Do not go ) = 1 − ∑ Pr (Go to k ) k =0 1 = 1 + exp[x t ' β + (1 − σ )I ] 10 z ' γ where I = ln ∑ exp kt . (1 − σ ) k =0 (3) (4) In the above equations, the i subscripts for the individuals are suppressed since the form of the model is the same for each individual in the data set. Some characteristics X could, in principle, vary across individuals. Here β denotes the parameter vector for characteristics that vary across individuals and or choice occassions but not across choices, γ is the parameter vector for choices that vary across choices, and (1 - σ) is the coefficient on the nested logit Inclusive Value. The nesting structure for the model of daily choices is straightforward and depicted in Figure 3. On any given open day, divers choose to go or not go fishing. If they choose to go fishing, they choose among the set of locations. For identification, we normalize the utility of not going to zero. Note, however, that the utility of not diving captures the utility of leisure, work opportunities outside of fishing, and the value of being avoiding exposure to unsafe diving conditions. Our strategy for estimating the model involved first drawing a random sample of 30 divers followed over the entire period. This relatively smaller sample of 27,000 choice occasion observations was used for preliminary specification testing. We used the smaller subset to determine the effects of using different backward lags for our expectations variables, different averaging methods for prices, and the impacts of various diver-specific variables computed from the data set. We also estimated several alternative specifications, including RPL models of just the location choice branch, that are ultimately not usable in our simulation exercise. In the final analysis, we estimated a parsimonious model, using the entire 401,151 observation data set, containing just the most important explanatory variables. The estimated parameters are shown in Table 5. Included among variables that are not presumed choice-specific are three coast-wide weather variables and a day of the week dummy variable. The weather variables include: WP (wave period), WS (wind speed), and WH (wave height), all computed as averages over the weather buoy data for the twelve hour period preceding noon of the day in question. The day of the week dummy (DWEEK) is one on Friday, Saturday, and Sunday, reflecting reduced propensity to dive on weekends and on days that precede the Monday closure of the Tokyo Central Wholesale Market. Variables that are location-specific are distance to the center of the patch in question (DISTANCE), expected revenue (ER) in each patch, and the Inclusive Value, computed from the lower level utility branch.16 As Table 5 shows, all variables 16 Some of the methodological issues associated with computing expected revenues are explored in Smith (2000). Expected revenues reported here are calculated as the product of expected price and expected catch per trip. The former is a rolling one-month backward looking average across the entire northern California 11 have signs as expected and all are highly significant as one would expect. Since σ is significantly different from zero, the inclusive value coefficient is less than one. This has two implications. First, it suggests that the model is consistent with stochastic utility maximization.17 Second, it suggests that variance of utility for participation and for location choices are different. This implies that choices across branches of the decision tree are less similar than choices within each branch of the tree. For example, choosing between patches 7 and 8 on a given day is more similar than choosing between 8 and not going fishing on that day. These estimated nested logit equations may be used to generate simulated short term effects of changes in economic variables, essentially assuming a time period short enough so that there is no induced movement of fishermen between ports or between Southern and Northern California. A short term increase in expected revenues anticipated in a particular patch k will have a spatial substitution effect as effort is drawn from other patches. However, the increase in expected revenues in patch k will also have a participation effect because fishing urchin becomes a more attractive use of time overall. This participation effect will cause more effort to be distributed over all possible patches. The upshot is that the own effect of an increase in expected revenues will be positive. The cross effects may be positive or negative, depending upon whether the participation effect outweighs the substitution effect. In general, the signs of cross effects will hinge on all of the variable values, and the substitution effect will be larger for patches that are visited more frequently (because Pr(k) is higher). Table 6 shows some of the computed short term elasticities of patch choice with respect to revenues. A 10% change in expected revenues off patch 8, for example, will result in a short term increase in total participation of 11.03%. This increase in effort will be distributed mostly to patch 8 which will experience a 6.2% increase, and in small amount averaging 0.5% to the other patches. In order to examine how a permanent or long-term closure might affect a fishery, it is important to understand that there would be a more elastic response than those computed for the daily port-specific nested logit model. The long term response would be larger as fishermen relocated to other Northern California ports that were adjacent to more open areas, or even to another region such as Southern California if similar closures were not in effect there. In order to examine port switching activity, we estimate some port-switching models that aggregate up both geographically and in the time scale of choices modeled. Since the urchin fishery is a licensed limited entry fishery, we express the models in term so shares of total licensed divers. In general, port switching and region switching entail different lumpy costs that will be viewed differently by different individuals. Divers with families or other attachments to regional infrastructure will view larger scale move opportunities differently than other more mobile individuals. Similarly, individual divers may differ in alternative income earning opportunities and earnings potential in urchin diving in each potential location. For all of these reasons, we would expect port shares to respond sluggishly to differences in expected earnings fishery, so that for each day the expected price is based on the average price for the previous 30.4 days. Expected catch is a patch-specific rolling one-month backward looking average. These proved to be bet fit in our preliminary specification tests. In addition, one-month averages are convenient for simulating the model in the next section. 17 McFadden (1979), Daly and Zachary (1979). 12 opportunities across ports and regions. A contemporaneous rent differential may not be enough to induce individuals to switch ports but if the differential widens or persists, we might expect switching response. A convenient way to model sluggish adjustment is with a partial adjustment framework. Let there be a long term equilibrium share of divers in each port that is a function of relative opportunities across all options. Then we can write: s *mt = f m (Π t , ..., Π t ; θ m1 ,...,θ mM ), m = 1 ,..., M , (5) as the equilibrium shares. We posit that actual shares adjust to differences between the equilibrium shares and last period’s actual shares. With an additive error term, this leads to: s mt = λs mt −1 + (1 − λ ) f m (Π t , ..., Π t ; θ m1 ,...,θ mM ) + ε mt (6) We estimate versions of this model under the assumption that the forcing equations are linear in expected monthly revenues.18 All alternative specifications involve some ad hoc assumptions; we choose this one because it allows us to impose sensible restrictions. For instance, this formulation allows us to impose a sensible restriction that long-run shares sum to one when there are no expected revenue differentials across ports. One can also impose symmetry restrictions such as that the increase in shares in one port due to own port expected revenue increases are balanced exactly by reductions in other ports due to that revenue change. Finally, a linear formulation is easiest to estimate and most tractable for our simulations. In order to estimate the port share equations, we must first generate a proxy for expected rents, since they are not observed across space and time. We use a structural approach that generalizes the famous Leslie closed population model (Hilborn and Walters, 1992). Suppose that biomass evolves over time according to the simple equation: X mt = X mt −1 − H mt + Rm (7) where X mt is unobservable biomass in patch m at time t, H mt is harvest, and Rm is mean recruitment of new additions to the biomass. If we define revenues to be: Rmt = pqEX mt (8) 18 This is a compromise between consistency and tractability for our simulation modeling. Alternative models transform the exogenous regressors into the unit interval. Examples include the logistic transform, used by Miller and Plantinga (1999) to estimate land use shares. One could also use an inverse logistic transformation of the right hand side with an additive error. 13 where q is a catchability scaling coefficient, p is price per unit harvest, and E is effort, we may, by recursive substitution arrive at the following equation for catch per trip ( ymt ): t −1 ymt = qE X m 0 − ∑ H mτ + tRm τ =0 (9) In reality, catch per trip might persist for some time beyond the decline of abundance because divers possess information about relative abundance within larger patches. The persistence of catch per trip might differ across ports because of differences in habitat and size of the fishable area. A generalization of (9) that allows some persistence is then: t −1 ϕ(L ) y mt = β 0 + β1 ∑ H mτ + β 2 t + ε mt τ=0 (10) This specification makes catch per trip a mean value, corrected for biomass drawdown with cumulative catch, and corrected for periodic mean recruitment with a time trend. The auto regressive structure helps purge residual autocorrelation, an especially important feature since the cumulative catch variable includes data that are lagged values of the dependent variable. We aggregated data to a monthly level, and corrected for heteroskedasticity by weighting each regression by the number of trips in each month. Table 7 presents GLS regressions with lags of catch per trip as additional regressors. We test for residual autocorrelation by regressing the residuals on right hand side variables and lagged residuals and performing joint hypothesis tests that the coefficients on the lagged residuals are zero. The test for residual autocorrelation is constructed so that the null is no autocorrelation, and we fail to reject the null in all but Crescent City, which has the fewest observations. The coefficients are nearly all significant and all have the expected signs consistent with the Leslie population modeling framework. We use the above GLS equations to generate projections of expected revenues per port and across Southern and Northern California ports aggregated together. Catch per trip forecasts are combined with last month’s mean port price to generate expected revenues. Since a good fraction of costs are determined with a share system, and since travels costs per trip are lumped into the constant for each port, anticipated revenues are a reasonable proxy for fishing profits or rents. These computed anticipated rents are used in explanatory variables in two port switching share systems. The first is a specification predicting the share of total harvesters choosing Northern and Southern California in each month, and the second is a Seemingly Unrelated Regressions system predicting share of Northern California fishermen in each of the six Northern California ports. We transform the expected revenue variables by converting them into logs and we impose restrictions on each specification to identify each model system. The main restriction imposed is a type of adding up restriction that ensures that the share increase resulting from a one percent rise in expected revenues in port k is identical to the sum of share losses in other ports associated with that increase. This is imposed by requiring that: 14 M ∑γ m =1 mk =0 In the SUR system, we also impose the restriction that the speed of adjustment is common within the system, suggesting that inertia against switching ports is the same for similar decisions, but not necessarily the same as with the North/South decision. With these restrictions, we drop one equation from each system and recover the parameters of the dropped equation computationally. Table 8 summarizes the results for the North/South system and Table 9 summarizes results for the Northern California ports. The North/South port share equation works well, explaining 82% of the monthly variation in the shares of fishermen participating in each of the large regions. The model suggests some moderate responsiveness to revenue differentials, tempered with a fairly sluggish adjustment process. The implied short run elasticity of shares with respect to changes in either region’s expected revenues is about 0.10, and the long run elasticity of response is about 0.33.19 The coefficient on the lag (0.861) suggests, for example, that it would take about a year and a half for shares to reach 90% of any new equilibrium induced by a change in relative expected revenues. The responsiveness to revenue changes is symmetric and an F test fails to reject the hypothesis that the coefficients are equal and opposite signed between the North and South regions.20 The SUR regressions also work well, although explaining much less monthly variation than the North/South model. The coefficient on lagged shares (0.436) implies that fishermen move between ports in response to revenue differentials within Northern California much faster than they move in response to North/South differentials. A change in permanent expected rents between Northern California ports would induce an adjustment that would be 90% completed in about three months, ceteris paribus. For the most part, coefficients on expected revenues are in accord with a priori beliefs. Three own-revenue coefficients are positive and statistically significant, one is positive and insignificant, and the other negative and insignificant. Many of the off-diagonals are negative and significant, but a couple have positive significant signs. The data on revenues across ports are fairly collinear and hence we expect that some of the aberrations are due to insufficient relative variation. In the next section we integrate the nest logit and seemingly unrelated shares models with a dynamic spatially explicit biological model to forecast the implications of reserve formation with behavioral response. V. An Integrated Bioeconomic Model of Reserve Formation In this section we outline a spatially explicit and dynamic bioeconomic model of the sea urchin fishery. The model is innovative in several respects. First, the model is a true bioeconomic model, integrating a population model of the urchin with a behavioral model of the harvesting sector so that the equilibria generated are joint bioeconomic 19 The short run elasticity is the gamma coefficient divided by the share and the long run elasticity is the short run elasticity divided by (1 – λ). 20 The test statistic for an F test is 0.0446, and the critical value is 3.93 at a 5% confidence level. 15 equilibria. Second, the biological model is explicitly spatial and dynamic. We depict the sea urchin population as a metapopulation of 11 discrete patches, each with its own natality/mortality and growth parameters. The populations are linked with a dispersal matrix capable of characterizing any type of qualitative dispersal pattern. We parameterize the dispersal matrix with parameters calibrated to mimic field observations of larval settlement along the Northern California coast. Third, the economic model also explicitly spatial and dynamic. We use the model discussed above to depict industry behavior as an aggregation of individual choices made by divers, each of which is presumed responsive to the relative expected profitability of participation and location. Finally, the economic model of harvester behavior and the biological model are linked and integrated over both time and space. This allows us to experiment with different spatially explicit policies, change economic and biological parameters, and trace out both short run impacts, long run steady state impacts, and the dynamic and spatial adjustments that take place in transition to steady states. A. The Metapopulation Model The metapopulation model developed to examine spatial management policies in the red sea urchin fishery consists of 11 discrete size-structured populations linked by a dispersal matrix.21 Each separate subpopulation has a size structure described by a von Bertalanffy equation, so that the size of an individual of age a in patch j is given by: ( Size j,a = L∞j 1 − e −k ja ), (11) where j is indexed from 0 to 10, a is indexed as a monthly time index from 1 to 360, and L∞j and kj are patch specific growth parameters. Note that L∞j is the terminal size of an individual organism, and this parameter dictates the maximum amount of biomass per individual. The model begins computations with a set of initial abundance matrices for each site. The initial abundance matrix in the first period is generally set with all zeros except for the month zero age class, which is read from an initial distribution file.22 The populations are then aged by advancing the abundance values for each month to the next older month so that A i,a = A i,a-1 where A denotes the number of organisms in the cohort. After the populations are aged, the numbers surviving in the population are computed, along with the catch. Survival is determined by a Beverton-Holt mortality relationship, which embeds both patch-specific natural mortality rates m j as well as fishing mortality rates f j if the size is above the minimum size limit L limit . We link the economic model of diver behavior to the population model by making monthly fishing mortality rates a function of predicted diver trips. Accounting for both natural and fishing mortality, survival of the number of individuals to age a becomes: 21 The metapopulation model is more fully described in Botsford et. al. (1999) In the simulations performed in the next section, we generally begin with initial conditions that simulate a non-harvested steady states, which would mimic the situation in 1988 when the northern California resource first came under exploitation. 22 16 A j,a A j,a e -m j = -m − f A j,a e j j if Size j ,a < Llim it (12) if Size j ,a > Llim it and total catch (C) consists of the sum of harvests of all sizes greater than the minimum size over all patches, which is 10 360 C = ∑∑ ( 1 − e −fj )w Sizeb j,a A j, a (13) , j =0 a =0 where w and b are allometric parameters relating weight and length. These parameters essentially convert number of organisms of each size to an aggregate measure of biomass. Note that for a given organism terminal size, L∞j , there is a corresponding terminal biomass. This, in turn, implies a maximum possible catch for a given number of organisms. The allometric parameters give rise to the possibility of a non-convex production technology whenever b>1, which is the common case. The metapopulation model also computes egg production, larval dispersal, settlement and survival. Egg production is computed after survival has been calculated for each month. If the month is a spawning month, then egg production in patch j is computed with: a = 360 ej= ∑ a =0 α x β A j,a Size j ,a where x = 0 if Size j ,a > Lmaturity if Size j ,a < Lmaturity . (14) This equation sums the egg production from each size class, where there is only positive production for sizes greater than the size at reproductive maturity. The exponent on the egg production parameter is greater than one, since egg production increases exponentially with size. After eggs are produced, they are distributed spatially over the system, using a dispersal matrix which can take on a number of different qualitative forms. During the months in which larval dispersal is assumed to take place, settlement of larva is calculated. For each month of the egg production period, a fraction of egg production is presumed to survive and this is distributed via the dispersal matrix from each of the patches to each individual patch according to: s in = pDe (15) This 11x1 vector gives the array of settlement associated with the array of egg production from the system, modified by the survival probability p, and distributed by the dispersal matrix D. If all of the patches cover all possible dispersal sites, then the rows of D sum to one. Beyond that possible restriction, D is general and can characterize a variety of dispersal mechanisms. One particularly important dispersal mechanism is uniform dispersal. This refers to situations in which the production of larvae in each location 17 redistributes uniformly over the entire system, often referred to a common larval pool assumption. The number that actually end up settling successfully is then assumed to follow a stock-recruitment function, namely: s inj out (16) s j = −1 a + c −1 s inj This specification enables the model to simulate various density dependent larval survival mechanisms in the system, all of which may be patch specific. Once the settlement is out calculated for any given site, the successful settlers ( s j ) become the next period’s age zero entry and the growth process starts again. Appendix A contains baseline values of the parameters that we used to calibrate the spatial biological model. The parameters are based on ongoing field work being conducted off the coast of northern California by colleagues investigating the sea urchin in a joint long-term research program.23 Raw data on growth increments and adult size distributions have been gathered in dive transects at several exploited and unexploited sites along the coast. These have been used to compute growth and natural mortality coefficients using maximum liklihood techniques. We have also been using larval counts from brush collectors that trap larvae on a continuous basis and correlating these settlement patterns with sea surface, wind, and current patterns on a fine time scale to understand disperal processes. The dispersal matrix used here is calibrated to reflect our best available understanding of current patterns and their impacts on larval settlement off the northern California coast. B. Simulating Spatial Policies: Spatial Response Elasticities To simulate the implications of spatial closures, we combined the spatially explicit biological model with the model of spatial behavior estimated with the logbook/landings ticket data base. The key link in the integrated model is the connection between monthly fishing mortality in each patch f jt and monthly trips, or: 4 f jt = (Trips jt )hq = (ot ∑ d p p pjt )hq ) p =1 where h is diving hours per trip, q is the scaling or catchability coefficient, ot is the number of choice occasions in that month24 , d p is the number of divers in port p , and p pjt is the probability of a trip from port p to patch j in month t. We presume h is fixed across all ports and time periods, and we use q to calibrate model forecasts to actual aggregate harvests. For short run projections, we fix the number of divers in each port and project 23 This work is reported in various sources, including Botsford et. al (1994), Morgan (1997), Smith et. al. (1998), Botsford et. al. (1999), Morgan et. al. (2000), and Wilen et. al. (2002). 24 Choice occasions are determined by the season closure regulations, generally the same across all northern California ports. 18 diver participation and spatial choices for each port using probabilities from the nested logit model. For long run projections, we allow the number of divers per port to be endogenous, using the SUR port share and North/South share results. In all simulations, since expected revenues are rolling one month backward averages, we use actual lagged catch per trip multiplied by an exogenous price to predict trips and shares of divers in each location. In the simulations reported here, we use a stylized and reduced-dimension model that focuses on four ports.25 Table 10 shows the implication of short and long term adjustments of effort to changes in revenues in terms of elasticities. Columns one and two take patch choice probabilities as given, and assume that there is no adjustment at the intra-port level. All elasticities are inelastic, not surprising given the infrequency of port switches. Column one shows, for example, that a 10% revenue increase in patch 8 would induce a 0.9% increase in trips to patch 8 in the very short run. Without further changes in divers in ports off patch 8, there would be an intermediate run adjustment involving patch switching toward patch 8 as well as some increase in participation, shown in column 2. The combined effects of these inter patch substitutions would raise the responsiveness to 1.78% Over the long run, a persistent increase in patch specific revenues would draw divers to nearby ports, as well as divers from southern California. Columns 3 through 5 show the increase in response elasticities associated with allowing inter-port and inter-region movement. With both port adjustment and full adjustment within and across patches in each port, elasticities rise to multiples of the short run elasticities. Long run elasticities with full adjustment are mostly in the unitary to elastic ranges, with differences reflecting the number of divers and trips in each patch. For example, the heavily fished areas in patches 7 and 8 show response elasticities of 1.014 and 0.876 respectively, whereas lightly fished patches 3 and 4 have elasticities of 1.588 and 1.784 respectively. C. The Importance of Behavior To understand how economic behavior matters to the forecasts of impacts of marine reserve formation, we compare simulations of the integrated bioeconomic model (the ECON model) with a standard biology only model that we dub the NOECON model. To keep things simple, we start with the less elastic model that forecasts patch choice and participation with the nested logit specification only, and not the port switching model. These models are closer to being identical, except that fishing mortality is made endogenous and dependent upon economic conditions in the ECON model. The NOECON model assumes constant total fishing effort before and after reserve formation and it is assumed (as typical in biological modeling) that effort is distributed uniformly. We incorporate a dispersal pattern that mirrors our current thinking about coastal oceanography off northern California. In particular, correlations of larval settlement in our collectors with weather events suggest that there are two gyres, one off Point Reyes (just south of Bodega) and one west of Point Arena. During upwelling events, larvae are swept south into the gyres and then are swept back and redistributed in weather relaxation 25 The two northern- and southern-most ports of Half Moon Bay and Crescent City are excluded, leaving Bodega, Point Arena, Fort Bragg, and Albion. Both excluded ports handle smaller amount of urchin. The reduced dimension port share system uses the estimated adjustment parameter for North/South choices, and incorporates symmetric off-diagonal coefficients for the substitute ports. 19 events. This combined advection/upwelling mechanism appears to collect larvae in the two pools and then deposit them non-uniformly, with more deposited close to the gyres and less deposited as distance increases. Figure 4 shows how the integrated model calibrates with the actual harvest path during the drawdown of the northern California urchin population over the past decade. We adjust the catchability coefficient to mimic actual harvests and then simulate the integrated model out to a steady state. We incorporate assumptions that mirror regulations in place, including seasonal and weekly closures and a minimum size limit of 3.5 inches. Table 11 shows the results of the simulations, including simulations of closures of patch 8, the heavily fished area off Fort Bragg.26 There are several important points to glean from this table about how incorporating behavior matters. First, the NOECON model dramatically overstates the extent of decline when calibrated to the drawdown phase of early fishery exploitation. The intuition for this is obvious once stated. Early in a fishery’s development, abundance is high and revenues are high, drawing large amounts of effort into the fishery. This results in a drawdown phase that ultimately overpredicts the amount of effort and fishing mortality that will remain after a steady state is approached. For example, the NOECON model predicts a steady state harvest of only 386 thousand pounds and a dramatically reduced egg production reflecting the assumed low reproductive potential of the overexploited biomass. In contrast, the ECON model continuously adjusts effort to profitability. As the fishery is drawn down, effort exits and fishing mortality falls, generating an ultimate steady state that is much larger. Under the ECON scenario, the steady state harvest is more than double the NOECON prediction, at 830 thousand pounds. Perhaps more importantly, the predicted egg production is almost five times that predicted with the NOECON model. The higher egg production emerges from two sources. First, since overall exploitation is lower, total reproductive biomass and egg production are larger. But second, the ECON model predicts a spatial distribution of effort that depends upon relative spatial profits. Areas that are high cost (eg. more distant) will be lightly exploited and hence will serve as de facto “reserves”, even without closures.27 Table 11 also shows how misleading biological models without behavior may be about the impacts of reserve formation. For example, the NOECON model calibrated to the approach path predicts that a closure of patch 8 will produce a 40% increase in steady state harvest, coupled with a 48% increase in system wide egg production. In the NOECON model, it is assumed that displaced effort adjusts immediately an in full numbers to the remaining open areas. Thus total harvest falls to zero in the closed area but increases in the open areas. Whether the loss over the whole adjustment period is made up by the corresponding gain depends on initial conditions and adjustment speeds. In this simulation, regardless of whether there is an initial net loss, it is compensated over the whole transition path because the present value of a closure is positive. Contrast 26 We use patch 8 to demonstrate the impacts of closures mainly because most literature suggests that reserves are likely to be beneficial to fisheries production in two instances: a) either when the closed patch is heavily exploited; or b) when the closed patch operates as a source rather than a sink. The gyre pattern and north-south pattern of larval flow makes patch 8 a type of source, although the definition is not always clear in the literature. 27 The steady-state spatial coefficient of variation for NOECON is 0.28 whereas for ECON it is 0.43. We know that spatial variation in the NOECON model must be due to net larval dispersal differences. In the ECON model, spatial variation is also due to differential fishing effort responding to differential rents. 20 these results with the ECON case, which makes the transition and ultimate steady state dependent upon relative profits. In the ECON simulation, a closure of patch 8 results in a 10% loss in steady state harvests, in addition to the transition losses associated with the closure. The result is that discounted revenues fall by 14%, rather than rise as predicted by the NOECON model. The importance of incorporating economic behavior into models intended to forecast the implications of reserves is thus profound. Importantly, the assumptions of uniformly distributed and unresponsive effort used in virtually all of the biological literature bias predictions toward overly pessimistic status quo harvest and egg production predictions, and overly optimistic predictions of harvest gains and the net economic costs of reserve formation. While calibrating a biological model incorrectly during the drawdown phase is one source of difference between NOECON and ECON predictions, it is not the only source. Suppose, in contrast, that a fishery that has adjusted to a steady state is modeled and calibrated to (lower) steady state effort levels. The third section of Table 11 shows what the NOECON model calibrated to steady state harvests would predict as response to closures. In this case, the NOECON model still predicts a steady state harvest gain, although much smaller than the approach path calibrated model. The smaller ultimate harvest gains are not enough to compensate for initial harvesting losses in the closed patch, however, and hence the system-wide present value of revenues falls. A more important difference may be in predicted egg production from the system. Because the NOECON model assumes uniform effort distribution, the whole system’s reproductive biomass is drawn down relatively uniformly to establish the status quo. In contrast, the ECON model predicts a heterogeneous distribution of effort reflecting distance from ports and other costs of remote patches. These act as defacto sources, hence contributing to predictions of overall egg production that are larger than the NOECON model. Figures 5 and 6 make clear how this arises. These figures show the steady state size distributions and egg production from a heavily exploited patch and from a closed patch. The closed patch has a wider age distribution with more larger individuals that exponentially produce more eggs in total. These depict how the de facto closures associated with relatively uneconomic patches can contribute in important ways to a system’s egg production. Table 12 shows that siting decisions may also look different, depending upon whether economic behavior is accounted for or not. This table shows the implications of siting reserves in various different patches in order to achieve different objectives such as harvest gains and system reproductive capacity gains. The NOECON model is calibrated to the same steady state harvest level as the ECON model, but with uniform effort distribution. Again, the ECON model predicts a drop in steady state harvests whereas the NOECON model predicts increases with certain patch closures. Decreases in harvests are predicted in the NOECON model from closures of the southern-most patches and in patches in the southern gyre around Point Arena. In contrast, in the ECON model, closing those patches is least costly in term of harvest lost, because they are most lightly exploited before the reserve. In terms of total system-wide egg production, the gains predicted by the ECON model are not as large proportionately as those predicted with the NOECON model. The NOECON model predicts relatively large egg production gains in patches 1 and 2, but mainly because the model overpredicts harvest pressure in those relatively unprofitable patches. The ECON model predicts relatively large egg 21 production gains from closing patch 2, but because it is lightly exploited and a large contributor to the first gyre. Large egg production gains are also predicted from closing patches 6, 7, and 8, but for different reasons. Patch 8 production gains come from closing a heavily exploited area, but patch 6 gains come from its critical role as a source feeding the second gyre. Overall, then, the rankings of best sites appears to depend upon both economic and biological dispersal. Ignoring economic dispersal seems to have critical effects on site choices, often missing the true configuration of the pre-reserve status quo, or mis-predicting eventual larval dispersal and harvest adjustments by failing to anticipate the behavioral response to closures. D. Port Switching and Long-run Behavior The simulations and comparisons between ECON and NOECON models reported above are mostly computed from steady states calibrated using a simple model of spatial behavior without port switching. In the long term, this fishery will tend toward an equilibrium reflecting balancing of economic opportunities across patches, across ports, and between northern and southern California. To understand how these more flexible opportunities to adjust in response to closures matter, we now examine a more comprehensive bioeconomic model that allows port switching behavior. The long run model is calibrated to reflect the long run regulatory goal of having 300 divers in the fishery. There is currently in place a limited entry program that induces attrition through minimum landings requirements and license fees. Our port choice models distribute a fixed number of total divers across the two northern and southern regions, and then across the ports in the northern region. We calibrate to a steady state that begins with about 130 divers making about 30 trips per year. The remainder of the 300 divers are assumed to locate in southern California. Table 13 shows how increases in mobility affect the steady state and changes associated with a closure of patch 8, with the first two rows repeating results from the previous section. By introducing port switching, the long run steady state begins with a smaller exploitation rate on the northern California stocks. The number of divers falls, although diving intensity is predicted to rise so that divers fish more days per season. Since the model without port switching is calibrated to current conditions including diver numbers close to 400, part of the difference in predictions arises out of the hypothesized long run reduction in number of about 25%. The remainder of the predicted reduction comes from divers rearranging their numbers to arbitrage opportunities across northern California ports, as well as switches to southern California. The model probably overpredicts switches to southern California because we have not endogenized the urchin population model to incorporate southern California biology. We thus parameterize returns in southern California to predict north/south shares. If divers actually did shift in large numbers, we would expect returns in southern California to fall, mitigating the profit differentials. In the third group of results in Table 13 we show how steady states in the northern California system would adjust with port switching. These are probably more realistic ranges, given the parameters of both the biological and economic models. Can reserves ever pay off in this system? The results of our simulations with and without port switching suggest that it is unlikely that reserves pay in this system from the perspective of the fishing industry. Adding more layers of spatial mobility does not 22 change the fundamental structure of the problem, either in the steady state or in transition. Our results confirm the analytical results of Sanchricio and Wilen (2002), namely that reserves are unlikely to increase aggregate harvests unless the fishery begins with a dramatically overexploited status quo. As they argue, there are two forces in play in those circumstances. First, with a highly overexploited reserve designate patch, the opportunity costs of closure are low. Second, as an overexploited population is rebuilt, it is more likely to have large payoffs in some dispersal systems (eg. density dependent migration). To further test these conclusions we experimented with two other situations designed to produce initial overexploitation. The fourth and fifth groupings in Table 13 show these simulations. In the fourth, we double ex-vessel prices and arbitrarily alter the participation equation of the nested logit almost four fold. This results in an overexploited system in the initial steady state, and as a result, a closure in patch 8 actually increases steady state harvests. However, the approach to the new steady state is slow, and the present value of revenues still drops as a closure is instituted. In the fifth set of results, we open movement up to port switching and region switching and perturb the southern California returns to generate high exploitation rates in the north. This scenario again fails to make reserves produce an economic return. VI. Summary and Conclusions This paper discusses the economic impacts of marine reserve creation, with particular attention to the role that economically motivated behavior plays in determining outcomes. We address a potentially important shortcoming of the vast marine reserves literature, namely the assumption that effort is fixed and uniformly distributed. Instead, we presume that effort in complex and realistic settings responds to economic incentives, particularly differential profit opportunities that are dynamic and spatial. To address the importance of behavior, we construct and estimate a series of spatial choice models using a comprehensive data base from logbooks and landings tickets. It is important to point out that these data are collected by fisheries scientists to guide regulatory decisions rather to aid economic research per se. Nevertheless, we show how this kind of data can also be used to examine economic choices in informative ways. Our spatial models address behavior across a range of time and spatial scales and at different levels of aggregation. We characterize the most important of these as short- and long-run models, the former taking place on fine (daily) scale and the latter taking place on longer (monthly and yearly) scales. The spatial models confirm what we would expect, namely that divers respond to differences in returns expected in different patches. The nested logit models show that fishermen respond negatively to weather risk and travel distance, and positively to expected returns, affected by both relative abundance and expected price. Although divers actually choose to participate only about 11% of the time, even a parsimonious model successfully explains a reasonable fraction of the variance in behavior. We link compliment the nested logit daily choice models with two systems of share equations designed to predict the share of total effort by port in northern California, and the share of divers locating in northern and southern California. Effort elasticities are reasonable and exhibit inelastic response in the short run to patch-specific expected revenue changes, and more elastic long run response allowing for port and region switching. 23 We link the spatial models of choice behavior to a biological model intended to capture the most important features of our case study. The biological model is cutting edge, and is one of the first to depict a multi-patch system with explicit hypotheses about larval dispersal. Most existing literature uses two patch models; ours is an eleven patch system calibrated with parameters derived from field data. A unique feature of the biological model is it incorporation of the “dual gyre” nature of coastal circulation in northern California. This feature represents the most current thinking about dispersal processes affecting urchin larvae and it adds detail and complexity that generate new hypotheses about reserve impacts and reserve siting questions. We simulate the model on a monthly time scale and the output from the model consists of aggregates such as system-wide egg production, total harvest, and current and present values of revenues, as well as patch specific system characteristics. Only a handful of papers consider the role of larval dispersal analytically; this paper is among the first to calibrate it empirically. Our results confirm that economic behavior is a critical determinant of the predicted impacts of reserves. Moreover, we show that the typical assumptions made by biologists for analytical tractability consistently bias the predicted impacts in a manner that makes reserves look more favorable than they actually might be. For example, a model that calibrates fishing mortality during a drawdown phase will consistently overpredict the extent of overexploitation compared with an economically based model that incorporates the natural decline in profitability as the steady state is approached. Since aggregate harvest is more likely to be increased in overexploited fisheries, results based on mis-calibrated fishing mortality are likely to be favorable. However, we also show that even without drawdown phase mis-calibration, the assumptions of uniform effort rather than economically motivated effort are likely to produce mistaken characterizations of reserve siting in realistic settings. We show, for example, that some remote or high cost or high risk areas are naturally less profitable and hence exploited less by fishermen. These patches form de facto reserves and hence contribute to overall egg production in ways not revealed by simpler uniform effort distribution models. The manner in which they contribute to both harvest and system reproduction is complicated, however, because their role depends upon economic, biological, and oceanographic factors. And the remoteness of particular patches is relative and dependent upon fishery independent factors such as port location and roadways. Our overall conclusions about the usefulness of reserves as a marine policy tool are at variance with the received wisdom in the biological literature for reasons discussed above. Our conclusions are more in accord with the very small amount of economic analysis that has been devoted to the topic. We find, as did Holland and Brazee in an early analysis, that reserves are more likely to produce harvest gains in an age structured model when the biomass is severely overexploited. We also find, as Holland and Brazee did, that even when steady state harvests are increased with a closure, the discounted returns are often negative reflecting slow biological recovery relative to the discount rate. We find results that confirm the Sanchirico and Wilen analytical work that show that easily exploited (low bioeconomic ratio) patches are most likely to be the best sites to produce both harvest and reproductive potential gains. Sanchirico and Wilen also have results pertaining to dispersal processes, but in a model in which (implicitly) only adults and juveniles migrate. The model used in this paper depicts a much richer and more realistic scenario in which dispersal is driven by oceanographic transport of larvae. Most 24 marine biologists believe that larval transport mechanisms are likely to be more important and more realistic determinants of dispersal in marine spatial systems than adult migration. We show that some of the notions about site selection currently in vogue appear to be based on naïve views of spatial dynamics that fail to incorporate fisher behavior. In particular, whether a particular patch is a source or sink depends very much on its relative level of exploitation as well as its physical placement in an oceanographic system. Patches with high intrinsic productivity in an unexploited system or a lightly exploited system may be less productive to the system as a whole when differential harvesting pressure (driven by relative economic opportunities) affects the spatial distribution of abundance and hence larval production. These results suggest, in contrast to the tone of recent dialogue among marine scientists about reserves, that there are still unanswered questions about whether they can deliver what is promised by simple modeling. At the very least, our integrated bioeconomic modeling efforts raise some new questions about whether oceanographic dispersal is the key driver of closure impacts, or whether harvester dispersal may be equally important. References Botsford, Louis W., Dale Lockwood, Lance Morgan, and James Wilen. (1999). “Marine Reserves and Management of the Northern California Red Sea Urchin Fishery”, CALCOFI Report no. 40, 87-93. Botsford, Louis W., James F. Quinn, Stephen R. Wing, and John G. Brittnacher (1993), “Rotatating Spatial Harvest of a Benthic Invertebrate, the Red Sea Urchin, Strongylocentrotus franciscanus,” in Proceedings of the International Symposium on Management Strategies for Exploited Fish Populations, University of Alaska Sea Grant College Program. Botsford, Louis W., Barry Smith, and James Quinn. (1994). “Bimodality in Size Distributions: the Red Sea Urchin Strongylocentrutus franciscanus as an Example”, Ecological Applications, 4(1), 42-50. Daly, Andrew and S. Zachary. (1979).”Improved Multiple Choice Models” in D. Henscher and Q.Dalvi (eds.), Identifying and Measuring the Determinants of Mode Choice, London, Teakfield, 335-357. Halpern, Ben. (2002). “The Impact of Marine Reserves: Does Size Matter?”, Ecological Applications (in press). Hannesson, Rögnvaldur (1998), “Marine Reserves: What Should They Accomplish?” Marine Resource Economics 13, 159-170. 25 Hastings, Alan and L. Botsford (1999). “Equivalence in Yield from Marine Reserves and Traditional Fisheries Management”. Science, 284. Hilborn, Ray and Carl J. Walters (1992), Quantitative Fisheries Stock Assessments: Choice, Dynamics and Uncertainty, New York: Chapman and Hall. Holland, Daniel S. and Richard J. Brazee (1996), “Marine Reserves for Fisheries Management,” Marine Resource Economics 11, 157-171. Kalvass, Peter E. and Jon M. Hendrix (1997), "The California Red Sea Urchin, Stongylocentrotus franciscanus, Fishery: Catch, Effort, and Management Trends," Marine Fisheries Review 59:2, 1-17. Kato, Susumu and Stephen C. Schroeter (1985), "Biology of the Red Sea Urchin," Marine Fisheries Review 47:3, 1-20. Ludwig, D. R. Hilborn, and C. Walters (1993). “Uncertainty, Resource Exploitation, and Convservation: Lessons from History”, Science, 260: 17-18. McFadden, Daniel (1978), “Modeling the Choice of Residential Location,” in Spatial Interaction Theory and Residential Location, Amsterdam: North-Holland. McFadden, Daniel (1979), “Quantitative Methods for Analyzing Travel Behavior of Individuals: Some Recent Developments,” in D. Hensher and P. Stopher, eds., Behavioral Travel Modeling, London: Croom Helm, 279-318. Miller, Doug J. and A. Plantinga. (1999). “Modeling Land Use Decisions”, American Journal of Agricultural Economics, 81, 180-194. Morey, Edward R., Robert D. Rowe, and Michael Watson (1993), "A Repeated NestedLogit Model of Atlantic Salmon Fishing," American Journal of Agricultural Economics 75, 578-592. Morgan, Lance. (1997). Spatial Variability in Growth, Mortality and Recruitment in the Northern California Red Sea Urchin Fishery, Ph.D. Dissertation, University of California, Davis. Morgan, Lance, Stephen Wing, Louis Botsford, Carolyn Lundquist, and Jennifer Diehl (2000). “Spatial Variability in Red Sea Urchin Recruitment in Northern California”, Fisheries Oceanography, 9(1), 83-98. Polacheck, Tom (1990). “Year Around Closed Areas as a Management Tool”, Natural Resource Modeling, 4(3), 327-353. 26 Quinn, James F., Stephen R. Wing, and Louis W. Botsford (1993), “Harvest Refugia in Marine Invertebrate Fisheries: Models and Applications to the Red Sea Urchin, Strongylocentrotus franciscanus,” American Zoologist 33, 537-550. Sanchirico, James N. and James E. Wilen (1999), "Bioeconomics of Spatial Exploitation in a Patchy Environment,” Journal of Environmental Economics and Management, 37: 129-150. Sanchirico, James N. and James E. Wilen (2002), "Bioeconomics of Marine Reserve Creation”, Journal of Environmental Economics and Management, 42(3). Smith, Barry, Louis W. Botsford, Stephen Wing. (1998). “Estimation of Growth and Mortality Parameters from Size Frequency Distributions Lacking Age Patterns: the Red Sea Urchin (Strongylocentrotus franciscanus) as an Example”, Canadian Journal of Fisheries and Aquatic Sciences 55, 1236-1247. Smith, Martin D. (2000), "Spatial Search and Fishing Location Choice: Methodological Challenges of Empirical Modeling," American Journal of Agricultural Economics 82(5), 1198-1206 . Wilen, James E., Martin D. Smith, Dale Lockwood, and Louis W. Botsford, (2002), Avoiding Surprises: Incorporating Fishermen Behavior into Management Models”, Bulletin of Marine Science (forthcoming). 27 Table 1 Number of Diving Trips by Northern California Port 1988-1997 Latitude Bin 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 18 21 25 Total HMB 3 572 18 45 28 2 3 BOD 32 1 32 62 748 596 6,055 752 39 9 16 19 1 PTA ALB CRC Total % of Trips from Port in Bin 3 282 3 604 19 77 93 750 599 6,475 5,920 7,877 6,674 13,172 6,705 2,643 217 113 39 18 291 0% 95% 0% 0% 0% 0% 0% 94% 0% 96% 0% 36% 98% 0% 0% 0% 0% 0% 97% 285 52,289 49% 3 2 394 5,002 7,556 333 104 41 30 1 10 47 151 5,416 4,692 56 19 2 671 FTB 8 1 8,370 13,466 10,394 13 119 131 914 8,360 6,589 2,593 217 111 39 15 19,101 Notes: The boxed numbers indicate that the port is located in that bin. There is no activity in bins 17, 19, 20, 22, 23, and 24. 28 Table 2 Diver Spatial Mobility Across Patches Number of Active Divers # of Patches Active In 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 50 44 20 10 4 1 0 0 0 0 0 60 51 19 8 0 0 0 0 0 0 0 146 72 59 36 20 6 2 0 1 0 0 139 72 60 32 23 12 8 4 3 0 0 139 62 52 36 38 13 15 3 0 0 0 99 65 46 32 18 15 9 0 0 0 0 107 67 45 25 13 13 2 0 0 0 0 49 48 30 21 15 3 1 1 0 0 0 37 44 25 21 0 4 0 0 0 0 0 36 45 33 22 7 3 0 0 0 0 0 43 33 27 18 5 3 1 0 0 0 0 43 39 24 9 9 1 0 0 0 0 0 Total Divers 129 138 342 353 358 284 272 168 131 146 130 125 Weighted Average # Patches 2.05 1.82 2.25 2.53 2.68 2.60 2.33 2.54 2.35 2.51 2.40 2.24 1 2 3 4 5 6 7 8 9 10 11 Table 3 Diver Spatial Mobility Across Northern and Southern Regions Number of Active Divers Year Total Divers Just S. Cal. Divers Just N. Cal. Divers Just One Region Divers Both Regions Divers Share in Both Regions 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 210 387 684 735 714 738 826 539 524 525 440 465 81 249 342 382 356 454 554 371 393 374 306 336 117 113 272 260 250 186 191 106 77 95 109 108 198 362 614 642 606 640 745 477 470 469 415 444 12 25 70 93 108 98 81 62 54 56 25 21 0.06 0.06 0.10 0.13 0.15 0.13 0.10 0.12 0.10 0.11 0.06 0.05 29 Table 4 Port Switches To ALB BOD CRC FTB HMB PTA SOC From ALB BOD CRC FTB HMB PTA SOC 24 3 328 14 194 129 26 3 1 324 60 10 3 68 31 175 233 11 6 Total 692 536 29 1206 19 32 8 6 9 525 278 6 9 193 182 7 527 7 Total 121 208 6 247 9 203 686 507 29 1184 70 1114 888 794 4478 233 72 1149 Table 5 Nested Logit Estimates Not Location-Specific Variable Coefficient Standard Error Z - statistic Constant WP WS WH DWEEK 1.06 -0.18 -0.11 -0.74 -0.74 0.048 0.005 0.003 0.011 0.012 22.21 -34.69 -36.69 -70.36 -60.02 Coefficient Standard Error Z - statistic -7.27 0.08 0.22 0.036 0.001 0.027 -203.72 65.17 8.34 Location-Specific Variable DISTANCE ER σ Log-likelihood Observations Pseudo R2 (1) 2 Pseudo R (2) -189,878 401151 0.21 0.81 Pseudo R2 (1) is based on the log-likelihood in a Conditional Logit Model with choice-specific constants. Pseudo R2 (2) is based on the log-likelihood of n*ln(1/J), where J = 12 possible choices. 30 Table 6 Nested Logit Cross-Revenue Elasticities Mid-week distance adjusted 1% Change in ERj j 0 % Change in Pk k 0 1 2 3 4 5 6 7 8 9 10 Net Change in PGO 1 2 3 4 5 6 7 8 9 10 1.515 0.130 0.117 0.119 0.117 0.124 0.126 0.129 0.130 0.126 0.104 0.042 0.598 0.045 0.046 0.045 0.048 0.049 0.050 0.050 0.049 0.041 0.085 0.093 1.212 0.094 0.093 0.098 0.100 0.102 0.103 0.100 0.083 0.080 0.097 0.088 1.145 0.088 0.093 0.094 0.097 0.097 0.094 0.078 0.086 0.104 0.094 0.095 1.221 0.099 0.101 0.103 0.104 0.101 0.084 0.065 0.080 0.072 0.073 0.071 0.938 0.077 0.079 0.079 0.077 0.064 0.058 0.071 0.064 0.065 0.064 0.067 0.836 0.070 0.071 0.069 0.057 0.047 0.057 0.051 0.052 0.051 0.054 0.055 0.672 0.057 0.055 0.046 0.043 0.052 0.047 0.048 0.047 0.050 0.051 0.052 0.620 0.051 0.042 0.058 0.071 0.064 0.065 0.064 0.067 0.069 0.070 0.071 0.837 0.057 0.112 0.136 0.122 0.124 0.122 0.129 0.131 0.134 0.135 0.131 1.580 2.736 1.063 2.163 2.051 2.190 1.675 1.491 1.196 1.103 1.492 2.857 Table 7 GLS Structural Model of Expected Catch Per Trip Coefficient SOC HMB BOD PTA ALB FTB CRC Intercept 96.82 (1.06) 3607.34 (8.7)** 2455.42 (7.55)** 913.60 (3.57)** 426.29 (2.55)** 1021.66 (4.52)** 1869.89 (2.14)** Cumulative Catch -9.07E-06 -0.006142 -0.000213 -9.32E-05 (-1.99)** (-3.81)** (-6.31)** ('2.78)** Time Trend 7.72 (1.86)* ymt-1 15.27 (1.49) -8.05E-05 -0.000073 -0.010859 (-2.09)** (-4.92)** (-1.72)* 12.90 (3.84)** 9.03 (2.21)** 4.71 (1.84)* 8.26 (4.67)** 28.32 (1.28) 0.61 (7.51)** 0.62 (7.85)** 0.66 (6.63)** 0.65 (10.91)** 0.87 (10.99)** 0.52 (4.16)** ymt-2 0.21 (2.24)** -0.23 (-2.97)** 0.04 (0.35) 0.09 (1.73)* -0.11 (-1.51) ymt-3 0.09 (1.05) Observations R-Square 143 0.916 71 0.414 123 0.871 116 0.888 116 0.856 122 0.943 34 0.838 F (for residual autocorrelation) F Critical 2.01 2.44 0.97 2.52 1.46 2.45 1.07 2.46 0.90 2.46 1.60 2.45 2.75 2.74 t-statistics are in parentheses under the coefficients. * indicates significant at the 10% level and ** at the 5% level. 31 Table 8 South/North Switching OLS Model of Port Shares Variable Parameter Coefficient t-statistic Constant α 0.028061 Lagged SOC Share λ 0.861212 17.242 ** ln(RSOC) γ SOC 0.056192 1.6678 * ln(RNOC) γ NOC -0.050767 -1.908 * 0.151 R2 0.8231 n 111 ** indicates significant at the 5% level and * indicates the 10% level. Table 9 SUR Results for Northern California Port Switching V a ria b le P a ra m eter C o efficien t t-sta tistic D iv er sh a re fo r: H a lf M o o n B a y α C o nstant HMB 0 .0 3 6 0 0 8 λ 0 .4 3 6 3 8 5 0 .7 1 1 0 .3 7 ** L agge d H M B S hare ln(R H M B ) γ H M B ,H M B 0 .0 0 9 2 9 5 ln(R B O D ) γ H M B ,B O D -0 .0 1 1 4 1 8 -1 .1 3 -0 .2 4 2 .2 9 ** ln(R P T A ) γ H M B ,P T A -0 .0 0 1 9 8 3 ln(R A L B ) γ H M B ,A L B 0 .0 0 2 1 1 0 .1 6 ln(R F T B ) γ H M B ,F T B -0 .0 0 6 5 7 3 -0 .5 7 ln(R C R C ) γ H M B ,C R C 0 .0 0 6 4 4 1 1 .4 0 B o d eg a α C o nstant -0 .5 8 8 0 5 8 HMB λ 0 .4 3 6 3 8 5 -4 .3 8 ** L agge d B O D S hare ln(R H M B ) γ B O D ,H M B 0 .0 0 5 3 9 2 0 .5 0 ln(R B O D ) γ B O D ,B O D 0 .0 2 0 2 2 6 0 .7 6 ln(R P T A ) γ B O D ,P T A 0 .0 3 8 5 6 2 1 .7 9 * ln(R A L B ) γ B O D ,A L B -0 .0 9 3 3 0 1 -2 .7 1 ** ln(R F T B ) γ B O D ,F T B 0 .1 4 3 5 6 9 ln(R C R C ) γ B O D ,C R C -0 .0 0 8 3 3 -- restric ted -- 4 .6 9 ** -0 .6 9 P o in t A ren a α C o nstant PTA 0 .5 2 4 7 7 7 λ 0 .4 3 6 3 8 5 2 .6 0 ** L agge d P T A S hare ln(R H M B ) -- restric ted -- γ P T A ,H M B -0 .0 1 8 3 -1 .1 5 ln(R B O D ) γ P T A ,B O D -0 .0 0 1 5 6 6 -0 .0 4 ln(R P T A ) γ P T A ,P T A 0 .0 6 8 0 6 4 ln(R A L B ) γ P T A ,A L B -0 .0 6 5 8 8 8 -1 .2 9 2 .1 3 ** ln(R F T B ) γ P T A ,F T B -0 .0 5 2 1 6 -1 .1 4 ln(R C R C ) γ P T A ,C R C 0 .0 0 0 8 5 1 0 .0 5 32 Table 9 SUR Results for Northern California Port Switching (cont’d) Variable Parameter Coefficient t-statistic Albion Constant α ALB 0.424906 2.99 ** Lagged ALB Share ln(RHMB) λ 0.436385 γ ALB,HMB -0.013528 -1.19 ln(RBOD) γ ALB,BOD -0.056048 -2.00 ** ln(RPTA) γ ALB,PTA -0.043831 -1.93 * ln(RALB) γ ALB,ALB 0.103488 2.86 ** ln(RFTB) γ ALB,FTB -0.067638 -2.10 ** ln(RCRC) γ ALB,CRC 0.033086 2.56 ** α FTB 0.15705 Lagged FTB Share ln(RHMB) λ 0.436385 γ FTB,HMB 0.019082 1.17 ln(RBOD) γ FTB,BOD 0.048374 1.17 ln(RPTA) γ FTB,PTA -0.064931 ln(RALB) γ FTB,ALB 0.047362 0.90 ln(RFTB) γ FTB,FTB -0.011484 -0.25 ln(RCRC) γ FTB,CRC -0.027462 -1.48 α CRC -- restricted -- Fort Bragg Constant 0.77 -- restricted -- -1.99 ** Crescent City Constant 0.008932 -- recovered -- Lagged CRC Share ln(RHMB) λ 0.436385 -- recovered -- γ CRC,HMB -0.001941 -- recovered -- ln(RBOD) γ CRC,BOD 0.000432 -- recovered -- ln(RPTA) γ CRC,PTA 0.004119 -- recovered -- ln(RALB) γ CRC,ALB 0.006229 -- recovered -- ln(RFTB) γ CRC,FTB -0.005714 -- recovered -- ln(RCRC) γ CRC,CRC -0.004586 -- recovered -- R2 (System Weighted) 0.357 n 555 ** indicates significant at the 5% level and * indicates the 10% level. 33 Table 10 Elasticities With and Without Port Model Trips to Patch j with Respect to Revenues in Patch j Elast. 1 Elast. 2 Elast. 3 Elast. 4 Elast. 5 Mean Revenues Farallons 0.251 1 0.098 2 0.207 3 0.017 4 0.126 5 0.119 6 0.057 7 0.111 8 0.090 9 0.122 10 0.233 0.508 0.197 0.457 0.068 0.301 0.271 0.127 0.239 0.178 0.241 0.461 2.076 0.807 1.607 1.520 1.483 1.066 0.974 0.775 0.699 0.983 1.895 2.328 0.904 1.814 1.537 1.608 1.186 1.031 0.886 0.788 1.104 2.127 2.584 1.004 2.064 1.588 1.784 1.337 1.101 1.014 0.876 1.223 2.355 1.5 x Mean Revenues Farallons 0.251 1 0.098 2 0.207 3 0.017 4 0.126 5 0.119 6 0.057 7 0.111 8 0.090 9 0.122 10 0.233 0.508 0.197 0.457 0.068 0.301 0.271 0.127 0.239 0.178 0.241 0.461 3.114 1.210 2.410 2.279 2.224 1.600 1.461 1.162 1.048 1.474 2.842 3.366 1.308 2.617 2.297 2.350 1.719 1.518 1.274 1.138 1.596 3.075 3.622 1.408 2.867 2.347 2.525 1.871 1.588 1.401 1.226 1.715 3.303 0.5 x Mean Revenues Farallons 0.251 1 0.098 2 0.207 3 0.017 4 0.126 5 0.119 6 0.057 7 0.111 8 0.090 9 0.122 10 0.233 0.508 0.197 0.457 0.068 0.301 0.271 0.127 0.239 0.178 0.241 0.461 1.038 0.403 0.803 0.760 0.741 0.533 0.487 0.387 0.349 0.491 0.947 1.290 0.501 1.011 0.777 0.867 0.653 0.544 0.499 0.439 0.613 1.180 1.546 0.601 1.260 0.828 1.042 0.804 0.614 0.626 0.527 0.732 1.408 Patch Elasticity Definitions 1) Short-run elasticity with no discrete adjustment. 2) Long-run elasticity with no discrete adjustment. 3) Elasticity with no port adjustment (short- and long-run are the same). 4) Short-run elasticity with both port and discrete adjustment. 5) Long-run elasticity with both port and discrete adjustment. 34 Table 11 Marine Reserves and Economic Behavior The Northern California Red Sea Urchin Fishery Steady-State Harvest (1,000 pounds) Steady-State Egg Production (Billions) Discounted* Revenues ($1000) 1,316 1,441 17,440 15,074 With Discrete Choice Behavioral Model (a=0.005) No Closure Close Patch 8 830 752 With No Economic Model - NOECON (a=0.005) Steady-state Calibration No Closure Close Patch 8 829 868 ** 434 553 17,400 16,423 With No Economic Model - NOECON (a=0.005) Approach Path Calibration No Closure Close Patch 8 386 545 *** 267 397 8,096 8,204 * Uses a 5% constant discount rate and assumes $1 per pound of sea urchin. ** Calibrated steady-state harvest to behavioral model. *** Calibrated approach path catch to actual catch. Table 12 Impacts of Dispersal and Spatial Behavior System-wide Totals for the NOECON Model System-wide Totals for the Discrete Choice Model SS Harvest (1,000 pounds) SS Eggs (Billions) SS Harvest (1,000 pounds) SS Eggs (Billions) 828.6 433.6 830.5 1316.0 Separate Patches Close Farralons Close Patch 1 670.6 749.4 566.5 580.3 819.1 827.1 1322.5 1317.5 First Gyre Close Patch 2 Close Patch 3 Close Patch 4 807.5 812.4 816.3 584.1 571.3 559.3 755.2 765.3 761.2 1395.4 1382.3 1385.3 Gyre Border Close Patch 5 911.6 559.7 755.0 1425.6 Second Gyre Close Patch 6 Close Patch 7 Close Patch 8 Close Patch 9 Close Patch 10 868.3 869.1 868.0 862.3 839.5 594.7 572.6 553.3 522.2 470.8 743.6 746.7 752.4 787.3 829.6 1446.9 1447.5 1440.6 1374.3 1316.0 No Closure 35 Table 13 Economics of Marine Reserves with Macroeconomic Shocks The Northern California Red Sea Urchin Fishery Steady-State N. California Divers Trips Per Diver Per Year Steady-State Steady-State Discounted* Participation Harvest Egg Production Revenues Rate (1,000 pounds) (Billions) ($1000) Discrete Choice Only (a = 0.005) No Closure Close Patch 8 131 131 29.9 25.3 13% 11% 830 752 1,316 1,441 17,440 15,074 57.8 47.2 25% 20% 638 576 1,627 1,692 13,400 11,660 16% 13% 802 728 1,399 1,495 16,846 14,683 46% 41% 883 921 720 879 18,548 17,362 796 910 20,402 18,865 Port Choice and Discrete Choice (a = 0.005) No Closure Close Patch 8 33 36 Port Choice and Discrete Choice (a = 0.005) - 50% Decrease in S. Cal. Revenues No Closure Close Patch 8 83 89 37.9 31.2 Discrete Choice Only (a = 0.005) Double prices and exogenous increase in participation rate No Closure Close Patch 8 131 131 107.1 96.5 Port Choice and Discrete Choice (a = 0.005) 75% Decrease in S. Cal. Revenues, double prices, and exogenous increase in participation rate No Closure Close Patch 8 56 68 174.5 143.1 75% 61% * Uses a 5% constant discount rate and assumes $1 per pound of sea urchin. 36 972 952 Figure 1 Diving Activity Coastal Histogram Trips in North-Central California 1988-1997 3,500 3,000 Number of Dives 2,500 2,000 1,500 1,000 500 0 37.95 Patch 1 38.10 38.25 38.40 Patch 2 38.55 Patch 3 Bodega 38.70 Patch 4 38.85 Patch 5 39.00 39.15 Patch 6 Pt. Arena 39.30 39.45 Patch 7 Patch 8 Albion Ft. Bragg Latitude Figure 2 Crescent City Northern California Sea Urchin Ports Ft. Bragg Albion Pt. Arena Bodega Half Moon Bay 37 39.60 Patch 9 39.75 39.90 Patch 10 40.05 Figure 3 Daily Participation and Location Choice Decision Do Not Go Patch 0 Patch 1 Patch 2 Patch 3 Diver i at time t Patch 4 Patch 5 Patch 6 Go Patch 7 Patch 8 Patch 9 Patch 10 Figure 4 Calibration of Bioeconomic Simulation Model 14,000,000 12,000,000 Total Catch (pounds) 10,000,000 8,000,000 6,000,000 4,000,000 2,000,000 0 1991 1992 1993 1994 1995 Actual Catch 1996 1997 Simulated Catch with a=0.005 38 1998 1999 Figure 5 Steady-State Size Distribution and Egg Production Patch 8 when Fished 450 6,000 400 5,000 350 4,000 250 3,000 200 150 Billions of Eggs Millions of Organisms 300 2,000 100 1,000 50 14 6 13 8 13 0 12 2 11 4 10 6 98 90 82 74 66 58 50 42 34 26 18 0 10 2 0 Size Class (mm) Organisms Below the Size Limit Organisms Above the Size Limit Egg Production from Size Class Figure 6 Steady-State Size Distribution and Egg Production Patch 8 when Closed 450 6,000 400 5,000 350 3,000 200 150 2,000 100 1,000 50 14 6 13 8 13 0 12 2 11 4 10 6 98 90 82 74 66 58 50 42 34 26 18 0 10 0 Size Class (mm) Organisms Below the Size Limit Organisms Above the Size Limit 39 Egg Production from Size Class Billions of Eggs 4,000 250 2 Millions of Organisms 300 Appendix A Baseline Parameter Values for Bioeconomic Simulations Parameter k m Linf Llimit Lmature f w b α β p a c Description Value growth natural mortality terminal size (mm) min. size limit (mm) min. size of sexually mature organism fishing mortality 1st allometric weighting parm. 2nd allometric weighting parm. 1st egg production parm. 2nd eggs production parm. survival probability resiliency settlement parm. carrying capacity settlement parm. 40 0.24 0.09 118 89 60 0.29 0.001413 2.68 5.47E-06 3.45 1.0 0.005 - 0.05 1.2E+07 - 2.4E+07