This paper introduces COMTE-LEFTIST, a hybrid explanation method for time-series classification, specifically applied to a Bitcoin recommendation system. The method combines COMTE, a counterfactual explanation framework, with LEFTIST, a feature-based local explainer, to enhance the interpretability of model predictions. COMTE-LEFTIST evaluates the impact of key shapelets from counterfactual examples on prediction scores, shedding light on how these shapelets influence trading decisions. We tested multiple models using one-minute Bitcoin closing price data across different time windows, with the MRSQM model achieving 70% accuracy for 30-minute sell/hold recommendations, outperforming deep learning models for shorter timeframes. This hybrid approach contributes to greater explainability in time-series classification, demonstrating how the integration of counterfactual and feature-based explanations can improve transparency in AI-driven financial strategies.