# How AI Berkshire Automates Portfolio Management and Rebalancing: A Complete Technical Workflow

> Discover how AI Berkshire automates portfolio management and rebalancing. Its technical workflow normalizes data, enriches market insights, and ranks positions by opportunity cost for optimized investing.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: automation
- Published: 2026-06-27

---

**AI Berkshire transforms raw holding data into concrete rebalancing plans through a multi-step workflow that normalizes inputs, enriches live market data, evaluates asset health, and ranks positions by opportunity cost.**

AI Berkshire treats portfolio management and rebalancing as a data-driven pipeline rather than an occasional manual task. According to the xbtlin/ai-berkshire source code, the core logic resides in [`skills/portfolio-review.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/portfolio-review.md), which orchestrates validation, analysis, and recommendation generation into a reproducible process. This workflow helps investors maintain concentration discipline, identify correlation risks, and optimize cash allocation.

## Input Normalization

The workflow accepts three distinct input formats defined in [`skills/portfolio-review.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/portfolio-review.md) (lines 26-33). Investors can submit holdings as percentage allocations, share quantities with prices, or reference a previously saved portfolio file. The system normalizes these disparate formats into a unified tabular holding model before proceeding to analysis.

## Live Data Enrichment and Validation

For each position, AI Berkshire launches background tasks to fetch current market data via WebSearch operations (lines 37-44). The `tools/financial_rigor.py verify-valuation` utility then classifies each holding's data richness as A, B, or C-grade, flagging low-confidence items that require additional scrutiny (lines 45-46).

## Single-Asset Health Diagnostics

Every holding undergoes a binary health assessment that evaluates current PE ratios, validates the original buying thesis, and assigns a "paper health" score (lines 49-55). The workflow answers three critical questions for each position: whether to buy today, hold for five years, and whether the investment thesis remains valid (lines 56-60).

## Portfolio-Level Risk Analysis

The system aggregates individual asset data into comprehensive risk metrics.

### Concentration Metrics

AI Berkshire measures portfolio concentration against Buffett- and Li-Lu-style targets, calculating the largest single holding's share, the top-three holdings' combined weight, total position count, and cash ratio (lines 65-72).

### Correlation and Theme Exposure

A cross-holding correlation matrix identifies hidden risks, such as regulatory exposure shared between Tencent and Kuaishou, while checking for currency or sector over-concentration (lines 78-86).

### Opportunity Cost Ranking

Positions are ranked by expected annualized return using `tools/financial_rigor.py three-scenario` calculations (line 106). The workflow compares the lowest-ranked holdings against cash-return benchmarks to identify capital reallocation opportunities (lines 94-108).

### Stress Testing

The pipeline runs qualitative scenario analyses for global recession, US-China conflict, interest rate spikes, and tech bubble conditions, estimating impact on individual positions and the overall portfolio (lines 113-122).

## Rebalancing Recommendations

Based on diagnostic outputs, the skill generates a concrete action table specifying *add*, *reduce*, *clear*, *new*, or *hold* directives for each ticker (lines 129-140). Each recommendation includes current percentage weights, target allocations, and a concise rationale grounded in the prior analysis.

## Cash Management Integration

When positions underperform cash returns, the workflow recommends cash allocation ranges of 10-30%, treating liquidity as a legitimate portfolio component rather than a temporary holding (lines 145-151). This reflects the Buffett principle that cash optionality provides strategic advantage.

## Report Persistence and Incremental Updates

All analysis results are rendered into a structured markdown report and saved to [`reports/portfolio-latest.md`](https://github.com/xbtlin/ai-berkshire/blob/main/reports/portfolio-latest.md) (line 174). This persistent file maintains the portfolio state for future incremental reviews and tracks historical rebalancing decisions.

## Invoking the Portfolio Review Skill

Users trigger the workflow through the command line interface. The following examples demonstrate the three supported input formats:

```bash

# Percentage allocation format

ai-berkshire /portfolio-review "腾讯30%, 美团20%, 茅台20%, 现金30%"

# Share quantity format with prices

ai-berkshire /portfolio-review "腾讯 500股 @480HKD, 美团 1000股 @130HKD, 茅台 200股 @1700CNY"

# Reference to saved portfolio

ai-berkshire /portfolio-review "我的持仓"

```

Each command triggers the full pipeline defined in [`skills/portfolio-review.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/portfolio-review.md), including validation via `tools/financial_rigor.py verify-valuation` and scenario modeling via `three-scenario`.

## Summary

- AI Berkshire normalizes percentage lists, share quantities, or saved files into a unified holding model via [`skills/portfolio-review.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/portfolio-review.md).
- The `tools/financial_rigor.py verify-valuation` utility grades data confidence while `three-scenario` calculates expected returns.
- Single-asset health checks evaluate PE ratios, thesis validity, and five-year hold potential.
- Portfolio-level analysis tracks concentration against Buffett-style targets, correlation matrices, and stress scenarios.
- Concrete rebalancing actions specify exact percentage adjustments with written rationales.
- Reports persist to [`reports/portfolio-latest.md`](https://github.com/xbtlin/ai-berkshire/blob/main/reports/portfolio-latest.md) for incremental future reviews.

## Frequently Asked Questions

### What input formats does AI Berkshire accept for portfolio management?

The system accepts three formats: percentage allocations (e.g., "腾讯30%, 美团20%"), detailed share quantities with prices (e.g., "腾讯 500股 @480HKD"), and references to saved portfolio files. All inputs normalize to a tabular holding model in [`skills/portfolio-review.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/portfolio-review.md) (lines 26-33).

### How does AI Berkshire determine which holdings to rebalance?

The workflow ranks positions by expected annualized return using `tools/financial_rigor.py three-scenario`, then tests the lowest-ranked holdings against cash-return benchmarks. It outputs specific actions—add, reduce, clear, new, or hold—based on concentration limits, correlation risks, and opportunity cost analysis (lines 94-140).

### What stress scenarios does the portfolio review workflow test?

The skill evaluates four qualitative scenarios: global recession, US-China conflict, interest rate spikes, and tech bubble conditions. It estimates qualitative impact on each position and the portfolio aggregate to surface hidden vulnerability correlations (lines 113-122).

### Where does AI Berkshire store the generated rebalancing reports?

The system writes structured markdown reports to [`reports/portfolio-latest.md`](https://github.com/xbtlin/ai-berkshire/blob/main/reports/portfolio-latest.md), which includes the overview, single-asset health tables, portfolio analysis, rebalance actions, and next-review schedule. This file serves as the persistent state for incremental updates (lines 152-166).