Ethereum Transaction Estimators Show Varied Accuracy by Gas Price

Coin WorldThursday, May 22, 2025 10:36 am ET
1min read

Ethereum transaction time estimates provided by Etherscan and EthGasStation have been a central tool for users of the platform, including ÐApp developers. EthGasStation is the most popular transaction processing time estimator, with popular wallets such as Metamask relying on it to recommend gas prices for desired processing speeds. Etherscan, one of the most popular Ethereum dashboards, has also developed its own processing time estimator. Despite their popularity and relevance, the accuracy of these estimators has never been empirically investigated.

To determine the accuracy of these estimators, four processing time predictions were obtained: two from EthGasStation (Gas Price API and Prediction Table API) and two from Etherscan (Pending Transaction page and the Gas Tracker). The actual processing time of each studied transaction was known, allowing for the calculation of the absolute error (AE) produced by each model. The models were then ranked based on their AE distributions using a three-step approach. This involved statistically analyzing the distributions, defining pairwise win and draw relationships, and ranking the models based on their win and draw relationships. The Alpha centrality was used to assign ranks to the models, with wins counting more than draws.

At the global level, the four prediction models were found to be equivalent, with roughly similar median AEs ranging from 40.8 seconds to 58.2 seconds. However, when stratifying the analysis by gas price categories, the Etherscan Gas Tracker webpage outperformed the others for “very cheap” and “cheap” transactions. The EthGasStation Gas Price API, on the other hand, outperformed the others for the remaining gas price categories. This means that if ÐApp developers wish to maximize prediction accuracy, they would need to use two models that belong to different online estimation services. If developers only care about the prediction of cheaper priced transactions, they can rely on the Etherscan Gas Tracker webpage alone. However, it is important to note that such a model is a black box, with Etherscan not providing any public documentation concerning the design of this model and how it operates.

In conclusion, while the four prediction models are equivalent at the global level, their accuracy varies when stratified by gas price categories. The Etherscan Gas Tracker webpage is the best model for cheaper priced transactions, while the EthGasStation Gas Price API is the best model for more expensive transactions. However, the lack of transparency in the Etherscan Gas Tracker webpage model is a concern for developers who wish to understand how the model operates.