An Algorithmic Framework for Systematic Literature Reviews: A Case Study for Financial Narratives

Battle Count: 71

The paper is primarily a systematic literature review methodology paper rather than a direct trading strategy paper. However, it synthesizes evidence that narrative-based indicators (sentiment, emotion, topic virality, network fragmentation) can outperform traditional indicators in forecasting asset prices, returns, volatility, and macroeconomic trends. The reviewed studies demonstrate practical applications in predicting stock returns, inflation, housing prices, and systemic risk. The framework itself could be applied to continuously scan and update literature on narrative-driven trading signals. Relevance is moderate as it provides the research foundation rather than implementable trading algorithms.

The SLR framework involves multiple stages: Scopus API querying, metadata filtering, transformer-based embedding generation, cosine similarity computation, Z-score standardization, KMO/CN diagnostics, PCA, and clustering with composite scoring. The pipeline is well-structured and uses standard Python libraries (HuggingFace SentenceTransformer, scikit-learn). However, it requires API access to OpenAI, Scopus subscription, and careful parameter tuning. The reviewed literature's methods range from simple dictionary-based approaches to complex transformer/GPT pipelines, varying significantly in implementation difficulty.

The framework is explicitly designed for reproducibility with predefined inclusion/exclusion criteria, systematic query construction, and algorithmic selection. However, it relies on OpenAI's proprietary text-embedding-3-small API (not fully reproducible without API access), Scopus database (subscription-based), and the data extraction phase remains manual. The clustering pipeline (PCA, K-means) is reproducible with open-source tools. The multilingual-e5-large-instruct alternative is open-source.