Scholarlink

<p>This study develops a many-objective portfolio optimization (MOPO) framework that integrates deep-learning forecasts with realistic investment constraints to generate executable long–short trading strategies. The framework leverages predicted three-month high and low price targets, adjusted by model-specific forecast errors, to construct conservative entry, stop-loss, and take-profit levels. Forecasting accuracy was benchmarked across nineteen advanced deep-learning architectures on thirty-five global equities, covering multiple sectors and regions to ensure robust justification. The hybrid deep-learning model consistently achieved the lowest average error of prediction (4–5 % for highs &amp; lows), whereas competing models often exceeded 10–40 %, confirming its reliability for this dataset while retaining a supportive role in the overall framework. The portfolio optimization is conducted on Indian equities, enabling realistic implementation under domestic market regulations, brokerage rules, and transaction costs. The MOPO model balances nine conflicting objectives: historical and predicted returns and risk, diversification entropy, skewness, kurtosis, risk-adjusted ratios, and budget adherence. Advanced evolutionary algorithms such as NSGA-II, NSGA-III, SPEA2, MOEA/D, NSDE-R, and aspiration-level-based NSDE-R are employed to generate Pareto-optimal portfolios that reconcile these trade-offs. Out-of-sample validation confirms that embedding forecast errors as explicit risk measures improves the reliability of stop-loss and target generation, leading to robust, investor-ready strategies. In addition, Statistical validation through ANOVA and effect size analysis confirms that algorithmic differences across key financial objectives are highly significant and practically meaningful. By strategically embedding predictive signals into a rigorously validated MOPO framework, this study contributes a robust and practically implementable methodology for investor-aligned portfolio selection in dynamic equity markets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence driven global equity many-objective portfolio optimization through hybrid deep-learning extreme-value forecasting and evolutionary algorithms

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Nikunj K. Joshi,
  • Jayesh M. Dhodiya

摘要

This study develops a many-objective portfolio optimization (MOPO) framework that integrates deep-learning forecasts with realistic investment constraints to generate executable long–short trading strategies. The framework leverages predicted three-month high and low price targets, adjusted by model-specific forecast errors, to construct conservative entry, stop-loss, and take-profit levels. Forecasting accuracy was benchmarked across nineteen advanced deep-learning architectures on thirty-five global equities, covering multiple sectors and regions to ensure robust justification. The hybrid deep-learning model consistently achieved the lowest average error of prediction (4–5 % for highs & lows), whereas competing models often exceeded 10–40 %, confirming its reliability for this dataset while retaining a supportive role in the overall framework. The portfolio optimization is conducted on Indian equities, enabling realistic implementation under domestic market regulations, brokerage rules, and transaction costs. The MOPO model balances nine conflicting objectives: historical and predicted returns and risk, diversification entropy, skewness, kurtosis, risk-adjusted ratios, and budget adherence. Advanced evolutionary algorithms such as NSGA-II, NSGA-III, SPEA2, MOEA/D, NSDE-R, and aspiration-level-based NSDE-R are employed to generate Pareto-optimal portfolios that reconcile these trade-offs. Out-of-sample validation confirms that embedding forecast errors as explicit risk measures improves the reliability of stop-loss and target generation, leading to robust, investor-ready strategies. In addition, Statistical validation through ANOVA and effect size analysis confirms that algorithmic differences across key financial objectives are highly significant and practically meaningful. By strategically embedding predictive signals into a rigorously validated MOPO framework, this study contributes a robust and practically implementable methodology for investor-aligned portfolio selection in dynamic equity markets.