From Preferences to Positions: Financial-Memory–Driven Portfolio Decisions

Author : Mohammad Hossein Soleimani

Abstract :This research introduces Intelligent, a ready-to-use, memory-centered multi-agent platform designed for comprehensive portfolio management. Its architecture orchestrates specialized agents—Research, Selection, Weighting, Rebalancing, and Quality Control—via a shared Financial Memory that logs investor preferences, past decisions, and market conditions, enabling transparent reasoning and self-correction throughout the process. The research details a year long comparison involving 10 historical client profiles, in which AI managed portfolios are developed alongside human benchmarks with the same goals and constraints. The AI performs daily, event driven rebalancing, while human portfolios are adjusted monthly. Risk-adjusted metrics assume rf = 2%. The research indicates that AI increases the average Sharpe ratio from 0.46 to 0.79, reduces volatility from 14.7% to 12.35%, and reduces the maximum drawdown from −14.6% to −9.4%. Significance is confirmed through paired sign tests (Return: 9/10, p=0.0107; Sharpe, Volatility↓, MaxDD↓, Treynor, Personalization: 10/10, p=0.00098; α=0.05). Ongoing monitoring with quick, constraint-aware adjustments seems key to improving risk management and personalization. Intelligent systems significantly reduce portfolio creation time from approximately 8 hours to 20 minutes, decrease decision latency from 3 hours to 2 minutes, expand bespoke coverage from 20 to nearly 10,000 simultaneous portfolios, and improve constraint adherence from 85% to 98%. These results demonstrate that a reasoning-focused, memory-augmented architecture can provide safer, more personalized, and highly scalable wealth management solutions, with statistically significant performance gains (p < 0.05).

Keywords :Intelligent, portfolio management, Selection, Weighting, Rebalancing, Quality Control.

Conference Name :International Conference on Global Business, Economics, Finance and Social Sciences (ICOGBEFSS -25)

Conference Place Dubai,UAE

Conference Date 4th Dec 2025