Scholarlink

<p>The effectiveness of portfolio construction is primarily dependent on the anticipated success of stock markets. The emergence of recent developments in machine learning has brought up significant opportunities for incorporating prediction theory into the practise of portfolio selection. However, numerous studies have demonstrated that it is not possible to get both very accurate predictions and substantial financial benefits by relying on a single prediction model. This work offers a new method for building a stock portfolio by combining the Firefly Algorithm (FA) enhanced AdaBoost model for forecasting stock returns with the Mean-Value-at-Risk (Mean-VaR) model for portfolio optimization. A new model predicts future stock returns using an AdaBoost regressor and hyperparameters adjusted via the Firefly Algorithm. It accomplishes this by examining historical data from NIFTY50 equities on the National Stock Exchange (NSE) of India spanning 2017 to 2023. Experiments reveal that the FA-AdaBoost model significantly reduces prediction error, reaching a 14.6% MSE drop relative to baseline AdaBoost and more closely matching projected returns with actual values. Expected returns guide the selection of a portfolio of 25 shares. Cardinality constraints to the Mean-VaR model help to optimize the portfolio. Although having lower downside risk, the most recently optimized portfolio outperforms conventional Mean-Variance portfolios in return with a 12.3% increase in the Sharpe Ratio. By means of swarm intelligence and ensemble learning, this approach demonstrates how essential it is to use risk-aware optimization to improve decisions in actual financial 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

A Hybrid Novel Approach for Stock Portfolio Construction

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Rajat Jaiswal,
  • Namita Srivastava,
  • Manoj Jha

摘要

The effectiveness of portfolio construction is primarily dependent on the anticipated success of stock markets. The emergence of recent developments in machine learning has brought up significant opportunities for incorporating prediction theory into the practise of portfolio selection. However, numerous studies have demonstrated that it is not possible to get both very accurate predictions and substantial financial benefits by relying on a single prediction model. This work offers a new method for building a stock portfolio by combining the Firefly Algorithm (FA) enhanced AdaBoost model for forecasting stock returns with the Mean-Value-at-Risk (Mean-VaR) model for portfolio optimization. A new model predicts future stock returns using an AdaBoost regressor and hyperparameters adjusted via the Firefly Algorithm. It accomplishes this by examining historical data from NIFTY50 equities on the National Stock Exchange (NSE) of India spanning 2017 to 2023. Experiments reveal that the FA-AdaBoost model significantly reduces prediction error, reaching a 14.6% MSE drop relative to baseline AdaBoost and more closely matching projected returns with actual values. Expected returns guide the selection of a portfolio of 25 shares. Cardinality constraints to the Mean-VaR model help to optimize the portfolio. Although having lower downside risk, the most recently optimized portfolio outperforms conventional Mean-Variance portfolios in return with a 12.3% increase in the Sharpe Ratio. By means of swarm intelligence and ensemble learning, this approach demonstrates how essential it is to use risk-aware optimization to improve decisions in actual financial markets.