Skip to content
Open access

A Risk-Aware Multi-Agent Decision Support System for Stock Trading on the Warsaw Stock Exchange: A Forecast-Combination Approach With Walk-Forward Evaluation

2026 · IEEE Access · Vol 14, pp. 152990-153013 · 0 citations · 51 references

Abstract

We study a multi-agent decision support system for stock trading on the Warsaw Stock Exchange (WSE), an underrepresented Central and Eastern European market. Five heterogeneous technical-analysis agents feed a transparent, confidence-weighted decision agent motivated by forecast-combination theory. The primary contribution is a rigorous and reusable evaluation framework that addresses four recurring weaknesses of backtested trading studies - in-sample parameter selection, absent transaction costs, neglect of risk-adjusted performance, and lack of significance tests - demonstrated through an in-depth empirical study of this ensemble. Evaluation combines anchored walk-forward tuning, calibrated commissions and slippage, Sharpe, Sortino, Calmar, and Martin ratios and the Ulcer Index, a volatility-matched comparison, bootstrap confidence intervals, regime decomposition, leave-one-agent-out ablation, and sensitivity analyses over decision-agent weights, selection objective, and costs, plus an empirical audit of the confidence heuristic. On 435 held-out trading days for WIG20, the ensemble trails passive buy-and-hold on total return in a bull-dominated window but improves every risk-adjusted metric and, once scaled analytically to benchmark volatility ex ante, total return as well; these differences are economically meaningful yet not statistically significant on a single sub-two-year window. Regime analysis attributes the profile to smaller losses in corrections and gains in sideways phases, offset by lower bull exposure. The same qualitative pattern appears in external validation on the FTSE 100 under UK-specific costs, and the ensemble remains competitive with LSTM and gradient-boosted-tree baselines under the identical protocol. The results support theory-motivated ensembles as risk-aware decision support rather than return enhancers.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.