Introduction

Traders see numbers: daily P&L +$150K, allocation changes, position moves. Investors want stories: why did the fund outperform today? What drove returns? Language models can automatically generate compelling narratives from structured market data, transforming raw numbers into investor-ready explanations. These narratives aid understanding and trust.

Narrative Generation from Market Data**

Data Input**

Provide: daily return ($150K), factor contributions (tech +$200K, financials -$50K), top positions, key trades, market context (Fed announcement, earnings, sector rotation). LLM generates coherent narrative explaining the day.

Causality and Logic**

LLMs learn causal relationships from training data. "Large tech exposure + positive AI news → outperformance" is a learned pattern. Models generate narratives that connect market events to portfolio outcomes logically.

Narrative Structure**

Opening**

"The fund returned 1.2% today, driven primarily by strong tech performance amid AI enthusiasm."

Detail**

"Our overweight in semiconductor stocks contributed $200K as manufacturing data beat expectations. Conversely, financial exposure detracted $50K due to rising mortgage rates."

Closing**

"Looking forward, we maintain our tech overweight given secular AI tailwinds while hedging rate sensitivity through floating-rate bonds."

Quality and Customization**

Tone Control**

Generate narratives in different tones: technical (for sophisticated investors), simplified (for retail), bullish, cautious. Prompt engineering controls tone without changing underlying facts.

Length Control**

Generate 1-paragraph summary for daily updates, 5-paragraph analysis for weekly reviews. Same data, different depth. Adapt to audience and format.

Case Study: Daily Investor Updates**

Fund publishes daily updates for 500 retail and institutional investors. Previously: analyst manually writes 3-4 paragraphs explaining the day. Time: 30 minutes/day. Quality varies (analyst's mood, complexity of day).

Current: feed market data and portfolio positions to LLM. Model generates narrative. Analyst reviews (5 minutes), makes tweaks if needed. Quality is consistent; coverage is complete (no day is skipped due to time constraints).

Factual Accuracy**

Data Grounding**

LLMs can hallucinate: invent facts not supported by data. Mitigate by: (1) providing only relevant data (don't mention a stock if it wasn't traded), (2) prompt engineering for accuracy ("Only mention the following stocks: ..."), (3) post-generation fact-checking (validate each claim against data).

Systematic Checking**

Automated validation: does narrative mention stock X? Check that X was actually traded. Does it claim X contributed $Y? Verify contribution matches. Fact-check flags inconsistencies.

Investor Communication Benefits**

Accessibility**

Narratives make market moves understandable to non-technical investors. Numbers alone are opaque; narratives provide context and rationale.

Engagement**

Investors reading a clear narrative about why their portfolio moved are more engaged and trusting than those seeing only numbers. Narratives justify portfolio management decisions.

Regulatory Considerations**

Narratives constitute marketing/advertising materials. They must be accurate, not misleading. Generated narratives are subject to same rules as hand-written ones. Don't make false performance claims. Disclose AI generation if relevant to trust.

Advanced Features**

Comparative Narratives**

Generate narratives comparing fund performance to benchmark: "The fund outperformed the benchmark by 50 bps due to our 2% overweight in tech..." Comparative framing is informative.

Counterfactual Narratives**

Generate "what if" narratives: "Had we not hedged interest rate risk, returns would have been 30 bps lower given the rate move." Counterfactuals explain portfolio decisions and risk management.

Conclusion**

AI-generated narratives transform daily P&L numbers into compelling stories. With proper grounding in data and fact-checking, generated narratives enhance investor communication and trust. For funds publishing frequent updates, narrative generation saves time while improving quality and consistency. Investors benefit from clearer understanding of portfolio drivers and decision rationale.