Scholarlink

<p>Accurately predicting financial trends amidst noisy data and shifting market regimes is essential for traders and investors. Traditional machine learning models often lack interpretability, while technical indicators offer clear signals but are not data-adaptive and apply uniform criteria across all instruments. To bridge this gap, we propose Moving Average Randomized Trees (MART), an ensemble method that embeds multiple exponential moving average features directly into decision tree split criteria. At each node, MART randomly samples exponential moving average span lengths, evaluates candidate splits based on their annualized geometric return gains, and selects the split that maximizes compounded growth. We backtested MART on nineteen financial instruments including equity indices, exchange traded funds, individual stocks, a commodity proxy, and a currency index over a two-year period. For each instrument, we compared annualized geometric return, sharpe ratio, and maximum drawdown against twenty benchmark strategies comprising standard technical rules, classical machine learning classifiers, and recurrent neural network models. MART consistently ranks in the top for all metrics, delivering higher average geometric returns and lower average drawdowns. Nonparametric statistical tests confirm these improvements are significant.</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

Moving average randomized tree

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Vinay Kumar Giri,
  • Rahul Goswami

摘要

Accurately predicting financial trends amidst noisy data and shifting market regimes is essential for traders and investors. Traditional machine learning models often lack interpretability, while technical indicators offer clear signals but are not data-adaptive and apply uniform criteria across all instruments. To bridge this gap, we propose Moving Average Randomized Trees (MART), an ensemble method that embeds multiple exponential moving average features directly into decision tree split criteria. At each node, MART randomly samples exponential moving average span lengths, evaluates candidate splits based on their annualized geometric return gains, and selects the split that maximizes compounded growth. We backtested MART on nineteen financial instruments including equity indices, exchange traded funds, individual stocks, a commodity proxy, and a currency index over a two-year period. For each instrument, we compared annualized geometric return, sharpe ratio, and maximum drawdown against twenty benchmark strategies comprising standard technical rules, classical machine learning classifiers, and recurrent neural network models. MART consistently ranks in the top for all metrics, delivering higher average geometric returns and lower average drawdowns. Nonparametric statistical tests confirm these improvements are significant.