Week Three — Foundation model vs Brownian motion. Kronos on five-minute BTC.
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

A recent test compared Kronos, a foundation model trained on global crypto data, with a traditional Brownian motion model for five-minute BTC price predictions. Brownian outperformed Kronos in out-of-sample testing, raising questions about the utility of complex models for short-term trading.

Recent testing shows that Kronos, an open-source foundation model trained on global crypto data, does not outperform a traditional Brownian motion model in predicting five-minute Bitcoin price movements in out-of-sample tests.

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

Researchers conducted a comprehensive comparison between Kronos, a model with over 25,000 GitHub stars and trained on data from 45 exchanges, and a geometric Brownian motion baseline. Using historical trade logs from Polybot’s simulated trades, they reconstructed market context for 497 BTC trades, applying each model to predict the probability of the price closing above the open at five minutes.

The evaluation focused on three metrics: Brier score, log-loss, and hypothetical profit & loss. Results indicated that Brownian motion slightly outperformed Kronos across these metrics, especially in out-of-sample testing. Specifically, in the last 249 trades, the difference in Brier scores was statistically insignificant, suggesting Kronos does not provide a meaningful edge over the simpler Brownian model in this context.

Why It Matters

This finding challenges assumptions that advanced, learned models automatically deliver better short-term trading predictions in highly efficient markets like Bitcoin. It raises questions about the practical value of complex models for real-time trading and highlights the robustness of traditional mathematical approaches like Brownian motion for certain applications.

Products Worth Considering

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black

Bose TV Speaker Compact Soundbar with Bluetooth and HDMI-ARC, Black

As an affiliate, we earn on qualifying purchases.

Background

Over the past two weeks, a paper-trading bot called Polybot has tested various strategies against Polymarket’s five-minute crypto markets, revealing that most ‘edges’ are mechanical artifacts that do not survive out-of-sample testing. The Brownian motion model, a 100-year-old mathematical assumption, has performed surprisingly well in this environment. Kronos, a recent development in foundation models for financial data, was introduced as a potential improvement but did not demonstrate a clear advantage in the recent tests. The study used open-source methodology to ensure transparency and reproducibility.

“Kronos does not outperform Brownian motion in out-of-sample tests for five-minute BTC predictions, suggesting traditional models remain competitive.”

— Thorsten Meyer AI

“Our results indicate that in highly efficient markets, complex learned models may not provide a significant edge over classical stochastic models.”

— Research team behind the study

What Remains Unclear

It remains unclear whether different configurations of Kronos, larger model sizes, or alternative training data could improve performance. Additionally, the real-time trading applicability of these findings needs further validation, as the tests were conducted offline and on historical data.

What’s Next

Next steps involve testing Kronos in live trading environments, exploring larger models or different training regimes, and assessing whether model improvements can translate into tangible trading advantages under real market conditions.

Key Questions

Why did Brownian motion outperform Kronos in these tests?

Brownian motion, a simple stochastic process based on geometric assumptions, aligns well with the efficient and noisy nature of short-term crypto markets, making it a surprisingly robust predictor in this context.

Can foundation models like Kronos be improved to outperform traditional models?

Potentially, but current evidence suggests that for short-term trading, the added complexity does not necessarily translate into better predictions. Future research may identify configurations or training methods that enhance performance.

Does this mean advanced models are useless for crypto trading?

Not necessarily. While they may not outperform simple models in this specific setting, they could offer advantages in other contexts, longer timeframes, or different market conditions.

Are these findings applicable to other cryptocurrencies or markets?

The study focused on Bitcoin and five-minute intervals; results may differ for other assets or timeframes. Further testing is needed to generalize these conclusions.

Source: Thorsten Meyer AI

You May Also Like

After the Paycheck: The Book I Wrote Because Nobody Else Would Tell the Truth About AI and Your Income

Thorsten Meyer AI says After the Paycheck is available as a serialized field guide and ebook on AI, wages and ownership.

A War Room for Your Next Idea: Inside IdeaClyst

Discover how IdeaClyst offers founders a local AI-powered war room for idea validation, risk assessment, and strategic decision-making, all on their own machine.

The citation. Why generative engine optimization rewards the same brand on the least stable ground.

New analysis shows that generative engine optimization tends to favor the same brand repeatedly, raising questions about fairness and stability.

U.S. Steel CEO says ‘golden share’ won’t curb reform strategy

U.S. Steel CEO David Britt states that the company’s proposed ‘golden share’ will not impede ongoing reform strategies, reassuring investors and stakeholders.