How to use the ETF Analysis agent skill (etf-analysis)

What the etf-analysis skill makes the agent do

An agent skill is a folder of instructions and reference material that an AI agent can load when a task matches it. This one is defined by a single file, SKILL.md, whose description covers ETF product screening, fee comparison, tracking error, liquidity assessment, strategy application and quantitative allocation for the Chinese market.

The file is written in Chinese. It contains reference text on ETF types and structures, definitions of core metrics, a selection method for funds tracking the same index, notes on core-satellite and sector-rotation allocation, and example Python functions.

• Explain ETF types: broad-based, sector, thematic, smart beta, commodity, bond, cross-border (QDII), money-market, plus ordinary ETFs, LOFs, feeder funds and leveraged or inverse ETFs.
• Calculate tracking error, information ratio, premium/discount rate, liquidity measures and total fee rate.
• Score and compare funds that track the same index.
• Suggest core-satellite or sector-rotation approaches, with sample functions for each.

How to install the etf-analysis skill

You can install the skill in two ways. Download the ZIP file for etf-analysis from the page where you found this guide. Then choose one of the options below.

Option A, personal skills folder: unzip the file into ~/.claude/skills/ so that the etf-analysis folder, containing SKILL.md, sits inside that directory.

Option B, project skills folder: unzip the file into the .claude/skills/ folder of a specific project, so the skill applies only to that project.

Option C, Claude apps: upload the ZIP under Settings → Capabilities → Skills.

How to trigger the skill with natural requests

You do not need a special command. The agent decides whether to load the skill based on the description in SKILL.md, so ask in plain language about the ETF topic you care about. Naming the index, the fund codes or the metric you want helps the match.

Good requests name the decision or calculation you need and provide the data. For example, you can ask the agent to compare several funds on the same index using their fees, tracking error and liquidity, or to explain what a premium of a certain size means for a specific fund.

• Compare the CSI 300 ETFs 510300 and another CSI 300 fund on fees, tracking error and daily turnover.
• Calculate the annualized tracking error from these daily returns for the ETF and its index.
• Explain the premium or discount of an ETF given its market price and IOPV.
• Show how a 0.5% annual fee compares with 0.15% over 20 years.
• Explain why a leveraged ETF should not be held for long-term allocation.

Compare ETFs tracking the same index step by step

The source recommends a five-step comparison for funds tracking the same index: screen by scale, compare fees, compare tracking error over one and three years, check liquidity, then assess the fund company. Scale thresholds in the file are rough guides: funds under 200 million yuan carry higher liquidation risk, and funds above 10 billion yuan have ample liquidity.

Tracking error is the main measure of how well an ETF copies its index. The file defines the daily figure as the standard deviation of the ETF return minus the index return, and annualizes it by multiplying by the square root of 252. For A-share broad-based ETFs, the file rates annualized tracking error below 0.2% as excellent, 0.2% to 0.5% as acceptable, and above 0.5% as poor.

Premium or discount is (market price minus IOPV) divided by IOPV. The file explains that arbitrage normally pulls a premium or discount back toward zero, but cross-border QDII ETFs with quota limits and sector ETFs with many suspended stocks can stay at a premium.

• Scale: funds under 500 million yuan are excluded in the example scoring step.
• Fees: total fee equals management fee plus custody fee plus index fee, excluding trading commissions, stamp duty and subscription or redemption fees.
• Liquidity reference: daily turnover above 100 million yuan for broad-based funds and above 20 million yuan for sector funds; bid-ask spread below 0.05% is rated high quality.

Use the sample code for fee drag and scoring

The skill includes example Python functions. The fee drag function compares how different annual fee rates reduce the end value of an investment over a given number of years, assuming a fixed index return. The file's example call uses an 8% index return over 20 years with fees of 0.2%, 0.5% and 1.5%.

The scoring function gives up to 100 points: 30 for scale, 25 for total fee, 30 for annualized tracking error and 15 for average daily volume. The file describes it as a simplified illustration. Its weights and thresholds are not backtested in the file, so treat the output as a starting point for your own checks, not a ranking to trade on.

• Required libraries for the example code: numpy, pandas and scipy.
• You must supply the data, such as prices, index returns, IOPV and Level 2 order book data. The skill does not fetch it.

Safety notes: scripts, network and limits

The skill runs no scripts and uses no network access, according to its listing. It provides reference text and code for you or the agent to read and adapt. Any calculation happens only when you or the agent run code on data you provide.

The file contains specific fee and product figures for a past year and rule-of-thumb thresholds that may be out of date. It gives no personal financial advice and no risk profile. Do not trade live money based on the sample scoring or rotation functions, which are simplified and not backtested in the file.

The file also states that leveraged and inverse ETFs suffer volatility decay and should not be used for long-term allocation. Short-term use is the only scenario it describes, and holding periods under one month are its reference point.

• Review any code the agent writes before running it on real data.
• Check current fees, quotas and index rules from the fund company or exchange before relying on figures from the file.
• Arbitrage based on creation and redemption requires qualification; the file gives a typical threshold of about 1 million units.