Complete Guide to the China A-Share Fear & Greed (Sentiment) Index
Index classification, archived 5.5-year in-sample research, and Python / AI Agent integration of the A-share fear & greed index — also known as the market sentiment / investor sentiment index. Archived research and API examples included.
Maintained by FusionIntel Labs · Product documentation and data interpretation
What is the A-Share Fear & Greed Index?
The A-Share Fear & Greed Index — also called the A-share sentiment index, market sentiment index, or investor sentiment index (in Chinese: 恐贪指数 / 情绪指数 / 情感指数 / 投资者情绪指数) — is a composite 0–100 gauge of the overall mood of the China A-share market. Values close to 0 indicate fear (broad pessimism, heavy selling pressure); values close to 100 indicate greed (over-optimism, eager buying). The reading is a sentiment measure; it does not directly measure the amount of buying or selling.
Three primary use cases: contrarian timing (be cautious when greed is extreme; pay attention when fear is extreme), position sizing (scale equity exposure with sentiment percentile), and cross-market style judgment (combine with sector-level fear & greed to track rotation). FusionIntel provides A-share macro, sector and individual-stock fear & greed. Coverage and observation dates are given by the current catalogs and responses.
Reading the index
The API provides a 0–100 fear & greed reading. Use the published metadata and data date when interpreting it. These endpoints do not expose the model’s complete factor weights or revision policy.
Current bands and archived research buckets
The current site-wide bands are 0–20 Extreme Fear, 21–40 Fear, 41–60 Neutral, 61–80 Greed and 81–100 Extreme Greed. Decimal cutoffs are 20, 40, 60 and 80, each included in the lower band. Read the current five-band guide.
Archived study: historical thresholds
The table and simulations below retain the original study thresholds of 15, 30, 70 and 85. They have not been recalculated under the current bands and do not define current sentiment labels.
The table below is real in-sample statistics: 1,320 A-share trading days from 2020-12-01 to 2026-05-15. For every day we record the fear & greed reading, then look forward 5 / 20 / 60 trading days to compute the SH index return. These figures are reproduced from an internal archived report; the underlying code and snapshot are not currently publicly downloadable.
| Bucket | n | 5-day mean | 5-day win rate | 20-day mean | 60-day mean | 60-day win rate |
|---|---|---|---|---|---|---|
| 0–15 historical bucket | 109 | +0.94% | 69.7% | +1.88% | +3.09% | 68.6% |
| 15–30 historical bucket | 201 | -0.04% | 53.7% | -0.08% | +1.01% | 62.6% |
| 30–70 historical bucket | 839 | +0.02% | 51.4% | +0.32% | +0.94% | 51.4% |
| 70–85 historical bucket | 136 | +0.17% | 50.7% | +0.29% | -0.71% | 43.3% |
| 85–100 historical bucket | 30 | -0.08% | 36.7% | +0.04% | -0.40% | 42.9% |
These archived averages describe this sample only. Overlapping observations and post-selected rules do not establish statistical significance or future returns.
The n column counts 5-day observations. Valid 20-day counts are 109 / 201 / 828 / 134 / 28; valid 60-day counts are 102 / 190 / 806 / 134 / 28 in table order. Overlapping signals are not independent samples.
Historical simulation: enter on fear < 15, hold 60 days
Translating the conditional pattern above into an executable rule: on any trading day where the fear & greed index closes below 15, model exposure using subsequent daily index returns and hold for exactly 60 trading days; if the signal re-fires while in position, extend the hold by another 60 days. Same sample period as above (2020-12-01 → 2026-05-15, 1,320 trading days), with 0.05% per-side transaction cost:
| Strategy | Total return | CAGR | Sharpe | Max DD | Time in market |
|---|---|---|---|---|---|
| Buy & Hold (benchmark) | +19.74% | 3.50% | 0.22 | -27.27% | 100% |
| Enter on fear<15, hold 60d | +30.67% | 5.24% | 0.39 | -26.30% | 71% |
| Historical rule: avoid readings >85 | +22.76% | 3.99% | 0.27 | -29.20% | 98% |
In this archived sample the 60-day rule has a higher annualized return than buy-and-hold. The comparison does not establish why this occurred or whether it will persist.
For questions about the sample and calculation assumptions, contact support@fusionintel.net. The tables alone are insufficient for independent reproduction.
The archived study also reports that two-sided threshold timing and dual-market confirmation underperformed buy-and-hold in this sample. These results should not be generalized into a trading rule.
Limits of the archived study
The historical tables above reproduce the archived research report, not a newly rerun test. Observations overlap and horizon-specific sample counts differ. The 60-day rule was selected using the same sample; the results do not establish out-of-sample performance.
Further research
Compare fear & greed with available price and valuation data using aligned dates. Any proposed combination rule needs its own reproducible evaluation. US sector and A-share individual-stock fear & greed require Pro+ Max; commodity and ETF fear & greed require Pro+ or Pro+ Max.
Use observations, not trading guarantees
State the sample, time window and missing data. A low or high reading alone does not establish a buying or selling opportunity.
API call examples
Pull the available observations in the 30d window of the A-share macro fear & greed index:
curl -H "X-API-Key: sk_your_key" \
"https://api.fusionintel.net/v1/feargreed/a_stock_macro/shi_feargreedindex?period=30d"
# Response:
# {
# "metadata": {
# "category": "a_stock_macro",
# "code": "shi_feargreedindex",
# "latest_date": "2026-05-15",
# "columns": ["date", "price", "feargreed_index"]
# },
# "data": [
# ["2026-05-15", 3368.21, 42.5],
# ["2026-05-14", 3354.07, 39.8],
# ...
# ]
# }Python:
import requests, pandas as pd
API_KEY = "sk_your_key"
url = "https://api.fusionintel.net/v1/feargreed/a_stock_macro/shi_feargreedindex"
r = requests.get(url, headers={"X-API-Key": API_KEY}, params={"period": "1y"})
data = r.json()
df = pd.DataFrame(data["data"], columns=data["metadata"]["columns"])
df["date"] = pd.to_datetime(df["date"])
df = df.set_index("date").sort_index()
print(df.tail())Using it in an AI Agent
FusionIntel ships two native AI Agent integration paths: Claude MCP Server and ChatGPT GPT Actions. After setup the agent can answer questions like this directly:
"How has the A-share fear & greed index moved over the past month? Which sectors are most bullish right now?"
Claude Desktop config:
{
"mcpServers": {
"fusionintel": {
"command": "npx",
"args": ["-y", "https://fusionintel.net/downloads/fusionintel-mcp-2.0.2.tgz"],
"env": { "FUSIONINTEL_API_KEY": "sk_your_key" }
}
}
}Full setup guide: Claude MCP Setup.
FAQ
Does the fear & greed index predict future returns?
FusionIntel fear & greed measures the sentiment state represented by its dataset. An extreme reading alone does not establish a future price direction, a reversal date or a return forecast. Historical comparisons require a stated sample and evaluation method.
How often is the index updated?
Publication is mainly daily and may vary by source. Check the observation date. The public endpoint /v1/public/feargreed/latest does not consume customer API quota; cache age is separate from the market data date.
Can I use the fear & greed API on the free tier?
Free includes 100 calls/day, A-share prices (30d) and macro fear & greed (up to 30d). Pro adds valuation and A-share/HK sectors up to 1 year; Pro+ adds commodity and ETF fear & greed up to 3 years; Pro+ Max adds US sectors and A-share individual-stock fear & greed with all available history. Compare plans.