Can agents make better decisions in turbulent markets?
This question defines our research agenda. The available studies address price forecasting, financial information extraction and explainability separately. They do not establish that agents make better decisions in cryptocurrency markets.
Forecasting and complexity
When does a more complex model improve a forecast? We compare a Wavelet–Transformer model with persistence, which uses the last observed price. The published Bitcoin test does not favour the more complex model.
Summary and citationFinancial language and agents
Can rephrasing and aggregating candidate answers improve verbatim extraction? VERSA studies causal spans in financial texts. Extracting a stated cause is different from establishing economic causality.
Summary and citationExplanations we can check
How can explanations from financial AI systems be traced and checked? VAXAI is an oral communication on this topic. No published paper or validated experimental claims accompany it.
Summary and citationOpen questions
- When does an agent add measurable value over a simple decision rule?
- How robust are decisions to changing market conditions and incomplete information?
- What remains after accounting for fees, slippage and computational demands?
- Can an explanation be checked independently of the model that produced it?
Doctoral research
Provisional thesis title
«Agentes Inteligentes Actuando en Mercados Caóticos»
The doctoral work at Universitat Jaume I connects these questions. The studies shown here contribute separate pieces; they do not constitute a completed evaluation of agents acting in markets.
Other research: computational gastronomy
We also study scent–colour relationships and interpretable representations of dishes. This is a separate line of work. It shares methods of representation and evaluation, without evidence that it improves market decisions.
Other research