Intelligent agents in turbulent markets

From chaos,
to comparison.

Can an agent make better decisions in a turbulent environment?

That is the question guiding our work. We study parts of the problem: forecasting, extracting information and explaining. Each new experiment must state its comparison and the conclusions it can support.

The field is a generative illustration of that question. It shows neither prices nor experimental results.

Intelligent agents in chaotic markets: hypotheses, evidence and limits.

Aldan Jay · Personal research website

Research

A result worth examining

Bitcoin · Algorithms

Does a more complex model help?

We combined a Transformer with multiscale sentiment signals and compared it with persistence: forecasting the last observed price.

Read the study

In the reported one-day Bitcoin test, the more complex model does not outperform persistence.

The study evaluates forecasting errors. It does not evaluate agents, trades or trading costs. A complete reproduction has not been verified.

Papers and communication

All publications

2026Financial language and agents
VERSA: Verbatim Extraction via Rephrasing and Self-Aggregation for Financial Causality
Aldan Jay; Rafael Berlanga; Yoelvis Moreno; Vicent Santamarta.
We study verbatim extraction of causal spans from financial texts in English and Spanish. VERSA combines specialised agents, question rephrasing and iterative aggregation. Development results improve on a single-pass baseline; some component tests remain preliminary estimates.
The 7th Financial Narrative Processing Workshop @ LREC 2026 · ELRA · 139–145 · Published
Summary and citation Read via DOI
2026Forecasting and complexity
Algorithmic Complexity vs. Market Efficiency: Evaluating Wavelet–Transformer Architectures for Cryptocurrency Price Forecasting
Aldan Jay; Rafael Berlanga.
We compare a Wavelet–Transformer model with persistence for Bitcoin price forecasting. In the reported one-day test, the more complex model does not outperform the last observed price. The result concerns this model and this test.
Algorithms · 19(2) · 101 · Published
Summary and citation Read via DOI

Let’s discuss the research.

We welcome collaboration on evaluation protocols, financial language, agent decision-making and independently checkable explanations. A useful starting point is a specific question, an available dataset and a simple comparison.

Contact

How do we decide in chaos?

From chaos,to comparison.

A result worth examining

Does a more complex model help?

Papers and communication

VERSA: Verbatim Extraction via Rephrasing and Self-Aggregation for Financial Causality

Algorithmic Complexity vs. Market Efficiency: Evaluating Wavelet–Transformer Architectures for Cryptocurrency Price Forecasting

Let’s discuss the research.

From chaos,
to comparison.