📈 economics

Price Dispersion in the Barcelona Fish Auction: A Probabilistic Interpretation of the Failure of the Law of One Price

By analyzing nearly 83,000 transactions from the 2008 Barcelona fish auction, this study demonstrates that the failure of the Law of One Price in perishable markets stems not from inefficiency but from the superposition of distinct seasonal and quality-based pricing regimes, which are best captured by modeling prices as species-specific probability distributions rather than deterministic functions.

Original authors: Luca Di Gennaro Splendore, Domenico Costantini

Published 2026-08-04

📖 8 min read🧠 Deep dive

CC BY 4.0

Original authors: Luca Di Gennaro Splendore, Domenico Costantini

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are standing in a bustling marketplace where everyone agrees that a "perfect" apple should always cost exactly the same amount. This idea, known in economics as the "Law of One Price," suggests that if you have two identical apples and no one is hiding secrets or charging extra fees, they should sell for the same coin. For a long time, economists tried to prove this law by looking for reasons why prices didn't match, blaming things like confused shoppers or travel costs. But what if the market isn't broken? What if the price isn't a single, stubborn number at all, but more like a weather pattern?

This is the world of "econophysics," a fascinating corner of science where researchers treat money and markets like they treat atoms and stars. Instead of asking "What is the price?", they ask "What is the shape of the price?" They look at how prices scatter, much like how raindrops fall in a storm. Some storms are gentle and predictable; others are wild and chaotic. By studying these patterns, scientists can learn if the market is behaving like a calm lake or a turbulent ocean. Understanding this matters because if we think prices are just random mistakes, we might try to fix them with the wrong tools. But if prices follow a specific, hidden pattern, we might realize that the market is actually working exactly as nature intended, just in a way we didn't expect.

The Great Fish Price Mystery

In 2008, researchers Luca Di Gennaro Splendore and Domenico Costantini decided to investigate this mystery in the most logical place they could find: the Barcelona wholesale fish auction. Imagine a giant, noisy room where fishermen bring in their daily catch and professional buyers bid on crates of fish. The auctioneer starts with a high price and slowly lowers it until someone shouts "Mine!" It's a perfect, high-speed game of economics.

The researchers grabbed a massive dataset containing 82,947 transactions from that entire year. They focused on ten popular species, from red shrimp to sardines. Their goal was simple: test the "Law of One Price." If the law were true, every kilo of the same fish, sold on the same day, to the same type of buyer, should cost the same. But when they looked at the data, they found something wild. On a single Wednesday in February, a kilo of red shrimp sold for €20 and another for €49.50. That's a difference of more than double!

The Old Way vs. The New Way

Traditionally, economists would look at this price gap and say, "Something is wrong!" They would try to find a reason, like "Maybe the buyer was in a hurry" or "Maybe the fish looked slightly different." They treated the price difference as "noise"—just random static on a radio signal that they needed to filter out to find the "real" price.

But these researchers had a different idea. They suggested that the price difference isn't noise. Instead, they argued that the price is like a roll of the dice, but a very special kind of dice. They proposed that fish prices aren't a single number; they are a probability distribution. Think of it like this: if you throw a ball at a target, it doesn't always hit the bullseye. It lands in a pattern. Sometimes it's a tight circle (low variation), and sometimes it's a wide scatter (high variation). The researchers wanted to see what shape that scatter took for different fish.

The Shape of the Market

They tested several mathematical shapes to see which one fit the fish prices best. Here is what they found:

The Log-Normal Curve (The Most Common Shape): For seven out of the ten species, including the fancy red shrimp and the common hake, the prices fit a "log-normal" curve perfectly. In plain English, this means the prices are formed by a chain of small, multiplying decisions. Imagine a buyer starts with a base price, then adds a little bit for freshness, a little bit for the day of the week, and a little bit for how hungry the restaurant is. When you multiply these small factors together, you get this specific, slightly lopsided curve. It's the "standard" way the market works for most fish.
The Gamma Curve (The Simple Adders): For two cheaper, bulk fish (red mullet and Mediterranean horse mackerel), the prices fit a "Gamma" distribution. This suggests a simpler process where prices are just the sum of a few parts, rather than a chain of multipliers. It's like building a tower with blocks: you just add one block, then another.
The "Weird" Cases (The Mixtures): The most exciting discovery happened with sardines and langoustines (a type of lobster). These fish didn't fit any single curve. The data was too messy. So, the researchers used a "mixture model," which is like realizing that the crowd in the room isn't just one group of people, but three different groups wearing different colored hats, all mixed together.

The Three Regimes of Sardines

For sardines, the mixture model revealed three distinct "regimes" or worlds happening at the same time:

The "Damaged" World (8.7% of sales): These are fish that are a bit beat up or sold at the very end of the day. They sell for a tiny fraction of the normal price (around €1.03 per box).
The "Normal" World (47.8% of sales): This is the standard, everyday fish sold between February and October (around €16.03 per box).
The "Peak" World (43.5% of sales): This is the super-fresh, high-demand spring season where everyone wants sardines, and prices jump to around €21.77 per box.

When you mix these three groups together, the average price looks weird and confusing. But once you separate them, the pattern becomes crystal clear. The same thing happened with langoustines, where the model found a hidden group of "sub-standard" lots that the official labels didn't catch.

Why the Price Changes (It's Not Just Mistakes)

The researchers also looked at why the prices moved around so much. They found that the "Law of One Price" fails for two very specific, structural reasons, not because the market is broken:

Time Travel (Seasonality): The price of fish changes drastically depending on the month. For hake, the price in October was 47% lower than in January. For red shrimp, the price in December was 106% higher than in January. This isn't a mistake; it's biology. Fish spawn and grow in cycles. You can't store fresh fish like you can store canned beans, so the supply changes with the seasons, and the price has to move with it.
Size Matters (Heterogeneity): A small hake is not the same "good" as a big hake. The researchers found that smaller fish could sell for 60% less than larger ones. This proves that even if the species name is the same, the product is different.

The Unavoidable Mystery

After accounting for the season, the size, the day of the week, and who was buying, the researchers still couldn't explain about 28% to 60% of the price differences. They call this the "irreducible stochastic component."

Think of it like this: Even if you know the weather forecast, the temperature, and the humidity, you still can't predict exactly where a single raindrop will land. In the fish market, there is a genuine, unexplainable randomness. It comes from the fact that every lot of fish is unique, every buyer has a secret private value for it, and the bidding happens in a split second. This randomness isn't a failure of the market; it's a fundamental feature of it.

The Big Takeaway

The paper concludes that the "Law of One Price" doesn't fail because the market is inefficient or because people are confused. It fails because the market is doing exactly what a market for fresh, living things should do. The price isn't a single number waiting to be found; it is a distribution—a cloud of possibilities.

For some fish, that cloud is a simple, smooth hill. For others, like sardines, it's a mountain range with three distinct peaks. By using these probabilistic tools, the researchers showed that we can actually see the hidden structure of the market—seasonal shifts, hidden quality groups, and the natural rhythm of the ocean—that a simple average price would completely miss. They didn't just find a new way to count fish; they found a new way to understand how the market breathes.

Technical Summary: Price Dispersion in the Barcelona Fish Auction

Problem Statement
The Law of One Price (LOP) posits that identical goods in a competitive market should trade at a single price. However, empirical tests across various markets consistently reject this prediction, finding substantial price dispersion. Traditional literature attributes this dispersion to market frictions such as search costs, information asymmetries, or transaction barriers, framing price variation as "noise" around a deterministic equilibrium.

This paper challenges that deterministic framing by examining the Barcelona wholesale fish auction (Confraria de Pescadors de Barcelona). The authors argue that for perishable goods, price dispersion is not merely unexplained residual noise but a structured, probabilistic phenomenon. The central problem addressed is whether the failure of the LOP in fish markets reflects market inefficiency or the superposition of structurally distinct price-generating regimes driven by seasonality, product heterogeneity, and biological supply cycles. The study seeks to characterize the statistical laws governing price dispersion and determine if prices are better described as realizations of specific probability distributions rather than deterministic functions of observable attributes.

Methodology
The study utilizes a complete administrative dataset of 82,947 transactions from the Barcelona fish auction covering the full calendar year of 2008. The analysis focuses on ten commercially significant species, accounting for 65.5% of the market's annual revenue. The methodology proceeds in three stages:

Unconditional Distributional Fitting: The authors test four candidate parametric distributions—Normal, Log-normal, Gamma, and Weibull—against the empirical price data for each species. Model selection is based on the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and the Kolmogorov-Smirnov (KS) statistic. The Log-normal distribution is tested based on Gibrat's law of proportional effect (multiplicative price adjustments), while Gamma and Weibull distributions are tested for additive or flexible error structures.
Finite Mixture Modeling: For species exhibiting non-standard residual structures (specifically high kurtosis or skewness) after single-distribution fitting, the authors employ finite mixture models. Using the Expectation-Maximisation (EM) algorithm, they estimate three-component log-normal mixtures. This approach treats the price distribution as a superposition of distinct latent regimes (e.g., seasonal or quality-based), where each component represents a specific price-generating process.
Conditional Log-Normal Regression: To decompose variance, the authors estimate conditional log-normal regressions ( $\ln P_{i,s} = \mathbf{x}_{i,s}^\top \beta_s + \varepsilon_{i,s}$ ). Covariates include month fixed effects (seasonality), size grade (quality), day-of-week, and a top-buyer indicator. This allows for the quantification of the "explained" variance versus the "irreducible" stochastic residual, testing whether the remaining variance is random noise or structured uncertainty.

Key Results
The analysis yields a consistent distributional taxonomy and reveals the structural mechanisms behind price dispersion:

Distributional Taxonomy:
- Log-normal: Seven of the ten species (including all red shrimp grades, hake, sea bream, and langoustine) are best described by a log-normal distribution, supporting the hypothesis that auction prices result from sequential multiplicative adjustments.
- Gamma: Two low-value bulk finfish species (red mullet and Mediterranean horse mackerel) are best fit by a Gamma distribution, suggesting an additive error structure where price formation involves the sum of independent quality components.
- Mixture Models: Two species, Sardine and Langoustine, require three-component log-normal mixture models. For Sardine, the mixture improves the AIC by 15,201 over the best single-distribution fit. The components correspond to distinct regimes: damaged/off-season lots (8.7% of transactions), normal-season lots (47.8%), and peak-season lots (43.5%). For Langoustine, the mixture reveals latent quality segmentation within a nominal grade.
Variance Decomposition: Conditional models explain between 40% and 72% of the log-price variance. The remaining 28% to 60% constitutes an irreducible stochastic component. This residual is not unstructured noise; it is species-specific and well-characterized, indicating that price formation contains inherent uncertainty that no deterministic model of observables can eliminate.
Mechanisms of LOP Failure:
- Temporal Non-Stationarity: The parameters of the price distributions shift systematically over time. For example, the location parameter for hake shifts by 0.94 log-units between January and October, effectively halving the median price. This reflects biological supply cycles interacting with inelastic demand, a structural feature rather than a market failure.
- Cross-Sectional Heterogeneity: Size grades (calibre) induce significant price penalties (up to 60% for smaller hake), violating the LOP's assumption of homogeneity.
- Regime Superposition: For mixture species, the LOP fails because the market is not a single entity but a collection of overlapping sub-markets (e.g., peak season vs. off-season) operating simultaneously.

Significance and Claims
The paper claims to offer a probabilistic reinterpretation of the Law of One Price's empirical failure in perishable goods markets. Its primary contributions are:

Reframing Dispersion: The study argues that price dispersion in fish auctions is a structural property of the market mechanism, not a deviation from equilibrium caused by inefficiency. The "failure" of the LOP is the result of the superposition of distinct probabilistic processes (seasonal cycles, quality tiers) rather than market friction.
Methodological Innovation: It demonstrates that finite mixture models serve as diagnostic tools to uncover latent market structures (such as seasonal regimes or hidden quality tiers) that conventional hedonic regressions and single-distribution analyses cannot detect.
Theoretical Extension: The findings extend the domain of log-normal and Gamma price distributions from financial and commodity markets to wholesale fish auctions. It suggests that for thin, perishable markets, the appropriate economic concept is not a single equilibrium price but a sequence of distributions or a mixture distribution.
Nature of Price: The paper concludes that in such markets, price is best understood not as a deterministic number but as a random variable with a well-defined distribution. The irreducible stochastic component reflects the genuine uncertainty inherent in auction mechanisms where lot uniqueness and private valuations prevent deterministic prediction.

The authors maintain modesty regarding generalizability, noting the analysis is confined to a single market and year. However, they assert that the analytical pipeline is reproducible and transferable to other lonjas (fish markets) and commodity markets characterized by perishability and heterogeneity.

Technical Summary: Price Dispersion in the Barcelona Fish Auction

More like this