🌿 ecology

Robustness and management performance of MSY reference points derived from the hockey-stick stock-recruitment model under structural uncertainty

This study demonstrates that while Maximum Sustainable Yield reference points derived from the hockey-stick stock-recruitment model tend to be biased, their integration with adaptive learning and precautionary measures enables robust fisheries management with comparable yields to those achieved using the true underlying model, even under conditions of structural uncertainty and limited data.

Original authors: Ichinokawa, M., Okamura, H.

Published 2026-03-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ichinokawa, M., Okamura, H.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are the captain of a fishing boat, and your job is to catch just enough fish to feed your crew today without running out of fish tomorrow. To do this, you need to know the "Goldilocks Zone": the perfect amount of fish to catch that keeps the population healthy and the catch steady. In the fishing world, this is called Maximum Sustainable Yield (MSY).

To find this Goldilocks Zone, scientists use a mathematical map called a Stock-Recruitment Relationship (SRR). This map tries to predict: If we have X amount of adult fish (spawners), how many baby fish (recruits) will be born next year?

The Problem: The Map is Fuzzy

The trouble is, we often don't have enough data to draw a perfect map. We might only have a few years of records, or the number of fish has stayed roughly the same for a long time. It's like trying to guess the shape of a mountain when you've only seen the bottom of it.

Traditionally, scientists used two main types of maps:

The "Ricker" and "Beverton-Holt" models: These are complex curves that assume if you have very few fish, they will reproduce super fast to bounce back (like a rubber band snapping back).
The "Hockey-Stick" (HS) model: This is a simpler, straight-line map that says, "If the fish population drops below a certain point, the number of babies stays flat or drops off a cliff. We won't guess what happens beyond the data we have."

The Study: Testing the "Hockey Stick"

This paper asks a big question: Is it safe to use the simple "Hockey-Stick" map when we don't have enough data, even if the "real" map is actually one of the complex curves?

The authors ran thousands of computer simulations, acting out different fishing scenarios over decades. They treated the complex curves as "Truth" and tested how well the Hockey-Stick map performed when used by mistake.

The Findings: The "Safe Guess" vs. The "Precise Guess"

Here is what they found, using some everyday analogies:

1. The Bias-Variance Trade-off (The "Safe Guess" vs. The "Precise Guess")

The Complex Maps (Truth): When you try to draw the complex curve with very little data, your guess wobbles wildly. One year you think there are 1,000 fish; the next, you think there are 10,000. This is high variance. It's like trying to aim a dart at a moving target in the dark; sometimes you hit, sometimes you miss by a mile.
The Hockey-Stick (The "Safe Guess"): Because the Hockey-Stick refuses to guess what happens outside the data it has seen, it doesn't wobble as much. It's low variance. It's like aiming at a target that is painted on a wall; you won't hit the bullseye perfectly, but you won't miss the wall entirely either.
The Catch: The Hockey-Stick is often biased. It might consistently guess the "Goldilocks Zone" is a little too high or too low. It's a "safe" guess that isn't perfectly accurate, but it's stable.

2. The Danger of "Overfishing" the Guess
When the fish population was already low (depleted), the Hockey-Stick sometimes made a dangerous mistake: it told the captain to fish harder than it should, thinking the fish would bounce back faster than they actually would. This is because the Hockey-Stick assumes the population is stuck at a "floor" and doesn't account for the complex "bounce back" ability of real fish.

3. The Solution: The "Safety Net" Strategy
The paper's most important discovery is that you can use the Hockey-Stick map, but you can't use it blindly. You need to add precautionary measures (safety nets):

The "Precautionary Factor": Instead of fishing at 100% of the recommended limit, fish at 80%. It's like driving 5 mph under the speed limit when the road is foggy.
The "Catch Cap": Even if the math says you can catch 1,000 tons, put a hard limit on the boat so you never exceed the estimated maximum.
Adaptive Learning: Update the map every few years. As you get more data, the Hockey-Stick can "learn" and switch to the more complex map if the data supports it.

The Verdict: A Pragmatic Compromise

The study concludes that while the Hockey-Stick isn't a perfect map, it is a pragmatic tool for when we are flying blind.

Without safety nets: Using the Hockey-Stick alone can lead to mistakes, especially for slow-growing fish (like large bottom-dwelling flounder).
With safety nets: If you combine the Hockey-Stick with a "slow down" factor and a "catch limit," you get results that are almost as good as using the perfect map. In fact, for fast-breeding fish (like sardines), this combination actually keeps the catch more stable and prevents wild swings in how much fish you catch each year.

The Big Picture

Think of this like navigating a ship in fog.

The Perfect Map is a GPS with a clear view of the stars.
The Hockey-Stick is a compass that only points "North" based on where you are right now. It might not point to the exact destination, but it keeps you from crashing into the rocks.
The Safety Nets are the life jackets and the rule to "slow down."

The authors are saying: Don't panic if you don't have the perfect GPS. If you use the simple compass (Hockey-Stick) but wear your life jacket (precautionary measures) and slow down, you can still navigate safely to a sustainable future for the fish and the fishermen.

1. Problem Statement

Fisheries management relies heavily on accurate Stock-Recruitment Relationships (SRR) to estimate Maximum Sustainable Yield (MSY) reference points (RPs), such as $F_{MSY}$ and $B_{MSY}$ . However, reliable estimation of SRR parameters is often impossible due to insufficient data contrast (i.e., a lack of variation in historical spawning stock biomass).

The Dilemma: Traditional SRR models (Beverton–Holt [BH] and Ricker [RI]) often fail or produce unrealistic extrapolations when data contrast is low.
The Alternative: The Hockey-Stick (HS) model (segmented regression) is widely used as a pragmatic alternative because it avoids extrapolation beyond observed data ranges.
The Gap: While HS is common in practice (e.g., used in ~50% of ICES category-1 stocks and ~68% of Japanese domestic stocks), its management performance, bias characteristics, and risks when the true underlying biology follows a different SRR (like BH) are not fully understood. Specifically, it is unclear how HS-derived MSY RPs behave under structural misspecification and insufficient data, and what management strategies can mitigate potential biases.

2. Methodology

The study employed a hierarchical simulation framework using age-structured population dynamics to address three main objectives:

A. Simulation Design & True MSY RPs

Models: Simulated population dynamics using three SRR types: Hockey-Stick (HS), Beverton–Holt (BH), and Ricker (RI).
Parameters: Tested across four representative Japanese species (varying in size, longevity, and habitat: sardine, mackerel, pollock, flounder), various steepness ( $h$ ) values, and recruitment variability ( $\sigma$ ).
Goal: Define and calculate stochastic MSY RPs ( $F_{MSY}$ , $SB_{MSY}$ , $MSY$) under equilibrium and stochastic conditions for all SRR types.

B. Estimation Experiments (Bias & Variance Analysis)

Scenario: Generated data from a "True" BH model under insufficient data contrast (constant stock levels at 0.5, 1.0, or 1.5 $SB_{MSY}$ ).
Estimation: Attempted to estimate MSY RPs by fitting the HS model to this BH-generated data (structural misspecification).
Comparison: Compared the bias and variance of HS-derived estimates against estimates derived from the correctly specified BH model.

C. Management Strategy Evaluation (MSE)

Setup: Extended simulations into a 40-year management period.
Harvest Control Rules (HCR): Tested various Management Procedures (MPs) combining:
- SR Selection Strategies: Always use BH (MP-BH), Always use HS (MP-HS), or Adaptive selection based on AIC (MP-AIC).
- Precautionary Measures: Catch capping (limiting catch to estimated MSY) and Precautionary Factors ( $\beta = 0.8$ or $1.0$).
- Thresholds: Different $SB_{trigger}$ levels.
Metrics: Evaluated performance based on stock recovery ( $SB_{safe}$ , $SB_{10}$ , $SB_{40}$ ), yield (relative to true MSY), catch variability, and initial catch reduction.
Statistical Analysis: Used Generalized Linear Models (GLM) to determine which factors (species, $h$ , $\sigma$ , management strategy) most influenced management success.

3. Key Contributions

Definition of Stochastic HS RPs: The study formally defines and calculates stochastic MSY reference points for the HS model, demonstrating that they are highly sensitive to recruitment variability ( $\sigma$ ), unlike BH/RI models where bias correction terms often dampen this effect.
Bias-Variance Trade-off Characterization: It quantifies the specific trade-off of using HS under structural uncertainty: HS-derived RPs exhibit higher bias but significantly lower variance compared to BH-derived RPs when data contrast is low.
Adaptive Management Framework: It proposes and validates a management procedure that combines adaptive model learning (switching between HS and BH based on AIC) with precautionary measures (catch capping and $\beta$ -factors) to mitigate the inherent biases of the HS model.
Species-Specific Precautionary Insights: It reveals that the effectiveness of precautionary measures (specifically catch capping) is dependent on life-history traits (generation time), being highly effective for pelagic species but less so for demersal species.

4. Key Results

A. Properties of HS-Derived MSY RPs

Stochastic Sensitivity: Under HS, $SB_{MSY}$ increases significantly with higher recruitment variability ( $\sigma$ ). This is because variability causes biomass to dip below the "hockey stick" breakpoint ( $SB_{HS}$ ), where no compensatory recruitment occurs, delaying recovery and reducing yield.
Range of $SB_{MSY}/SB_0$ : The range of $SB_{MSY}$ relative to virgin biomass is much wider under HS (0.15–0.64) compared to BH (0.15–0.42) or RI. This suggests standard proxies (e.g., 30–40% $SB_0$ ) are unreliable for HS.
Variance: HS estimates have lower variance than BH estimates, making them more stable but potentially systematically biased.

B. Estimation Biases (Misspecification)

When the true SRR is BH but HS is fitted to low-contrast data:
- Bias: HS tends to overestimate $F_{MSY}$ (leading to conservative management) in depleted stocks and underestimate it in high-biomass stocks.
- Variance: The variance of these estimates is lower than when fitting the correct BH model to the same low-contrast data.

C. Management Performance (MSE)

Naïve HS (MP-HS/N): Without precautionary measures, HS-based management leads to persistent bias and slower stock recovery in depleted scenarios, though it maintains catches near MSY.
Adaptive + Precautionary (MP-AIC/C): The strategy combining adaptive model selection (AIC) and catch capping performed best.
- It gradually reduced bias as data accumulated and the correct BH model was selected more frequently.
- It achieved stock recovery and yield levels comparable to the "perfect" scenario (MP-BH/N).
- It significantly reduced catch variability and initial catch reductions compared to MP-BH/N.
Role of Capping: Catch capping was crucial for pelagic species (short generation time) to prevent TAC overshoots but had minimal effect on demersal species (long generation time).
Precautionary Factor ( $\beta=0.8$ ): Applying a 0.8 factor to target fishing mortality was broadly effective across all species, directly reducing realized fishing pressure.

5. Significance and Implications

Pragmatic MSY Implementation: The study validates the use of the HS model as a viable "entry point" for MSY-based management in data-limited fisheries where traditional models fail.
Robustness through Precaution: It demonstrates that while HS-derived RPs are biased, this bias can be managed. Combining HS with adaptive learning and precautionary measures (capping and $\beta$ -factors) creates a robust management system that limits yield loss and variability.
Policy Guidance:
- Managers should not rely on HS-derived RPs in a "naïve" manner.
- Catch capping is recommended for fast-growing, high-variability species.
- Precautionary factors (e.g., $\beta=0.8$ ) are universally beneficial.
- Standard biomass proxies (like 30% $SB_0$ ) should be re-evaluated when using HS, as $SB_{MSY}$ can be much higher.
Future Direction: The findings support the transition toward adaptive management frameworks that acknowledge structural uncertainty and utilize model averaging or switching strategies to improve long-term sustainability.