Predicting market efficiency from early orderbook signals

Researchers have built a framework that estimates how efficiently a market allocates resources using only the bids, asks and realized prices visible in its orderbook, without ever seeing the true reservation values of buyers and sellers. The work, published in EPJ Data Science by Thomas Asikis and Heinrich H. Nax, addresses a long-standing blind spot in market analysis: in real marketplaces, the numbers traders post rarely reveal their genuine maximum willingness to buy or minimum willingness to sell, so an economist observing the orderbook alone has little to say about actual performance.

The authors invert the standard logic of double auction experiments. Traditionally, experimenters assign participants induced reservation values and then measure how close the resulting trades come to the theoretical optimum, which yields a direct measure of allocative efficiency. The new approach runs that pipeline backwards. It trains predictive models on experimental market data so that, given only the observable orderbook stream, they can estimate key market features relevant to efficiency, particularly allocative efficiency, as early as possible in a trading session.

The challenge is considerable. There is no established model of strategically optimal behavior in double auction markets, and orderbook data are highly unstructured, non-stationary and non-linear. To tame this complexity, the researchers propose quantile-based normalization techniques that convert raw orderbook signals into inputs suitable for general predictive models. These normalized features draw on the underlying supply-demand structure of the market, allowing the models to learn patterns that transfer across different market conditions.

With those features in place, the team developed and trained several model classes, including linear regressions and gradient boosting trees. The models were evaluated on their ability to predict allocative efficiency from the earliest bids and asks posted in a market, before any trade has occurred, and then again as realized prices accumulate. According to the study, the models can predict allocative efficiency with reasonable accuracy from those first orderbook entries, and their predictions improve as additional price realizations become available.

Performance varies by prediction target and by market type, the authors report, so no single model dominates every setting. That variation matters for anyone hoping to deploy such tools in practice, because digital marketplaces differ widely in structure, participant composition and trading dynamics. The paper does not claim a single universal predictor; instead it documents which modeling approaches work best under which conditions, using controlled experimental data where the true efficiency benchmark is known.

The practical motivation is large in scale. Digital marketplaces processing billions of dollars annually represent critical infrastructure in sociotechnical ecosystems, yet their performance optimization lacks principled measurement frameworks that can inform algorithmic governance decisions regarding market efficiency and fairness from complex market data. A tool that estimates allocative efficiency from observable trading activity alone could give regulators, platform designers and researchers a way to assess market health without privileged access to participants’ private valuations.

Because the framework needs only bids, asks and price realizations, it holds significant potential for application to real-world market data, offering insights into market efficiency and performance even prior to any trade realizations. In a live marketplace, that means an analyst could form an early estimate of how well the market is likely to allocate goods simply by watching the orderbook unfold, then refine that estimate as trades execute. The authors emphasize that predictions improve with more realized price data, so the estimate becomes sharper over the course of a session.

The study is grounded in experimental markets rather than field data, which is both its strength and its limitation. Experimental control is what makes the inversion possible: only in the lab are true reservation values known, so only there can models be trained against a verified efficiency target. The authors present the framework as a step toward principled measurement, with the quantile normalization and the trained models forming a pipeline that could later be tested on observational data from actual marketplaces, where ground truth is unavailable and validation would need different methods.

For the science of market design, the result reframes what can be inferred from passive observation. Orderbook data, often treated as too noisy and strategic to support welfare analysis, can under the right normalization and modeling choices carry enough signal to estimate allocative performance. As algorithmic trading and automated market governance expand, tools of this kind could inform decisions about efficiency and fairness using the same data streams platforms already collect, turning the orderbook from a record of activity into an early diagnostic of market quality.

Subject of Research: A framework for predicting market allocative efficiency from orderbook data in double auction experiments

Article Title: An ‘inverse’ experimental framework to estimate market efficiency

Article References: Asikis, T., & Nax, H. H. (2026). An ‘inverse’ experimental framework to estimate market efficiency. EPJ Data Science. https://doi.org/10.1140/epjds/s13688-026-00684-9

Image Credits: AI Generated

DOI: 10.1140/epjds/s13688-026-00684-9

Keywords: market efficiency, orderbook data, double auction, allocative efficiency, quantile normalization, gradient boosting, experimental economics, algorithmic governance, digital marketplaces, predictive modeling, inverse, experimental