Where the crypto Fear & Greed index comes from

A number between 0 and 100 sits at the top of half the crypto dashboards on the web, labelled Fear or Greed, and it looks like a measurement. It is a recipe: a handful of other numbers, each itself computed by somebody, each rescaled against its own recent past, multiplied by weights the publisher chose and added up. There are several such recipes, under one name, and they do not agree.

This page is about how those recipes are built. It is not about what a reading means for anybody's positions. Both main publishers' pages go on to say what they think a high or a low reading implies for the market; that is their claim, it is not part of how the number is made, and nothing here repeats it or tests it.

How it works

A sentiment composite has the same skeleton as any other index — inputs, weights, a schedule — and the construction side of that is already covered in how a crypto index is built. What makes this kind different is what goes in. A price index combines prices of the same kind. A Fear and Greed index combines quantities of different kinds — a volatility, a volume, a count of posts, a market-cap share, a search-interest figure — which share no unit and cannot be added as they stand.

So every published version does three things before it can add anything up:

1. Computes each input, from some source, over some window. Each input is a derived number with its own methodology, usually somebody else's.
2. Turns each input into a score, by comparing it with a reference — its own average over a lookback, typically — so that "high" and "low" mean something on a common scale.
3. Weights the scores and sums them into 0–100, then cuts that range into named bands.

Every one of those steps is a choice, and the composite is only as reproducible as the least published of them.

Alternative.me: six inputs, published weights

The index that started the genre is alternative.me's. Its page, read on 27 October 2026, describes a daily index "for bitcoin only", built from six inputs with stated weights:

• Volatility, 25%. "The current volatility and max. drawdowns of bitcoin", compared "with the corresponding average values of the last 30 days and 90 days". An unusual rise counts toward fear. The page does not say how "current" volatility is measured, from which price, or how the 30-day and 90-day comparisons are combined.
• Market momentum and volume, 25%. "The current volume and market momentum", against the same 30- and 90-day averages, put together into one value; high buying volume in a rising market counts toward greed. The page does not say whose volume. That matters more than it sounds: why two sites report different volume for the same exchange covers the half-dozen choices that make two honest volume figures for bitcoin disagree, and the index inherits whichever set of choices its source made.
• Social media, 15%. Twitter analysis is running; a Reddit sentiment analysis is described as "still not in the live index". The method is to "gather and count posts on various hashtags for each coin" and measure how many interactions they receive, how fast, "in certain time frames". The hashtags, the time frames and the formula are not published.
• Surveys, 15%. Marked "currently paused". The description is of weekly polls on strawpoll.com — which the page discloses alternative.me also owns — with 2,000 to 3,000 votes each. The page does not say whether the paused 15% is redistributed to the other inputs, held at a fixed value, or something else.
• Dominance, 10%. Bitcoin's share of total crypto market capitalisation; a rising share counts toward fear. Dominance is a ratio whose denominator is whatever list of coins the market-cap source tracks, so it depends on which aggregator supplied it, and the page does not say.
• Trends, 10%. Google Trends data "for various Bitcoin related search queries", especially the change in search volume and the related queries Google suggests.

Two details of the page are worth reading closely. It introduces the inputs as "the five following sources" and then lists six, whose weights add up to 100. And it says that "each data point is valued the same as the day before in order to visualize a meaningful progress in sentiment change", which it does not explain further.

The bands are not on the methodology page at all. The API attaches a label to every value, and in the 3,187 daily values it returned on 27 October 2026 — our own reading, not a published rule — the labels fall at Extreme Fear up to 25, Fear 26–46, Neutral 47–54, Greed 55–75 and Extreme Greed from 76. Every value carries a timestamp of 00:00 UTC, the series starts on 1 February 2018, and four days are missing from it: 14 to 16 April 2018 and 26 October 2024.

CoinMarketCap: five inputs, unpublished weights

CoinMarketCap publishes a separate index with the same name and a different recipe. Its page lists five components:

• Price momentum: the price performance of the top 10 cryptocurrencies by market capitalisation, stablecoins excluded, assessed "relative to each other and the broader market".
• Volatility: the Volmex implied volatility indices BVIV and EVIV for bitcoin and ether — a forward-looking 30-day expected volatility derived from options prices, not the realised volatility of past prices that alternative.me uses. See implied volatility for the difference.
• Derivatives: the put/call ratio in the bitcoin and ether options markets. The page does not say whether the ratio is of open interest or of volume, over which venues or which expiries.
• Market composition: the Stablecoin Supply Ratio, "the ratio between Bitcoin's market capitalization and that of major stablecoins". Which stablecoins count as major is not stated.
• CMC proprietary data: "social trend keyword searches and user engagement metrics". The page says nothing more about where these come from than that they are proprietary, which is also what places them out of any outside reader's reach.

No weights are given. The page describes "multiple data sets and a complex weighting system", says the index "has been backtested thoroughly and is regularly reviewed", and publishes neither the weights, nor the normalisation, nor the backtest, nor a revision history.

The bands are published, in the API reference rather than on the index page: Extreme Fear from 1 to under 20, Fear 20 to under 40, Neutral 40 to under 60, Greed 60 to under 80, Extreme Greed 80 to 100. The latest-value endpoint is documented as updating every 15 minutes; the historical endpoint returns one value per day, at 00:00 UTC.

Same name, different numbers

Because the bands differ, one value carries two labels. A 45 is Fear on alternative.me and Neutral on CoinMarketCap; a 57 is Greed on one and Neutral on the other; a 22 is Extreme Fear on one and Fear on the other.

The values differ as well. Reading both APIs on 27 October 2026 — our own comparison, reproducible with the commands at the foot of this page — the two series overlap on 1,215 days, from 29 June 2023 to 26 October 2026. Over those days they move together (a correlation of 0.90) and still sit an average of 7.3 points apart, carry different labels on 41.5% of days, and differ by 20 points or more on 3.9% of them. The widest gap was on 13 April 2026: 12, Extreme Fear, on alternative.me against 54, Neutral, on CoinMarketCap. That is what two recipes that share almost no input produce.

A third publisher shows the same thing from the other direction. CoinStats publishes a methodology with the same six inputs and the same weights as alternative.me — volatility and volume against 30- and 90-day averages, Twitter hashtags, weekly surveys of 2,000 to 3,000 votes, dominance, Google Trends — updated every 12 hours, and it does not say the surveys are paused. Its number is its own: on 27 October 2026 it read 42, Fear, against alternative.me's 48, Neutral, and CoinMarketCap's 43, Neutral, and it labelled the previous week's 51 as Greed, which alternative.me's bands would call Neutral. Publishing the same weights does not produce the same index; the inputs underneath the weights are where the numbers come from.

Each input is somebody else's methodology

The composite is a weighted rank of several other opinions, and each of those has a method of its own. Three of them carry most of the uncertainty.

Social volume. "Social volume" is not one quantity. Santiment defines it, in its metric documentation updated 26 August 2026, as "the total number of social data text documents that contain the given search term at least once" — a post mentioning bitcoin ten times counts once — across 4chan, Telegram, Reddit, Twitter, BitcoinTalk and YouTube videos, matched for an asset on a search term built from its name, ticker and related terms with known false matches removed, in five-minute intervals. It also says that a change to labels or sources recalculates the previous month's data, and that data can be supplemented within a 12-hour period, so a figure read today can differ from the same day's figure read next week. LunarCrush defines social volume as public posts mentioning a topic across X, Reddit, YouTube, TikTok, Instagram and news, counted after topic classification and spam and bot filtering, with a rolling 24-hour figure as the common view; its sentiment is a 0–100 score weighted by engagement, so a widely shared post moves it more than an ignored one. Two vendors, two definitions of the same words, and neither Fear and Greed publisher says it uses either of them.

Search interest. Google Trends is not a count of searches. Google's own FAQ says the data is a sample of searches, that each point is divided by total searches in its place and time range and then scaled from 0 to 100, that low-volume terms show as 0, and that noise is added for privacy. A Trends figure is already normalised against the window it was requested for before an index normalises it again, and alternative.me does not name the queries it requests.

Implied volatility. BVIV is an index in its own right, built by Volmex from option and futures data consolidated across derivatives venues into one option book and smoothed by exponentially weighted averaging. Its page states that "index data is licensed". A component that is a licensed third-party index is a component the composite's reader cannot inspect except by buying it.

The scale is the last layer. Alternative.me scores volatility and volume against their own 30- and 90-day averages, so its index measures change against the recent past rather than a level: the same volatility reads as fearful after a quiet month and unremarkable after a turbulent quarter. CoinMarketCap does not say what it normalises against. In either case a reading of 25 is a statement about how the inputs compare with their own history, through a lookback the publisher chose, and not about any absolute state of the market.

What it costs

The indices themselves cost nothing to read. Alternative.me's API needs no key; its rules allow commercial use "as long as the attribution is given right next to the display of the data", forbid using the data to impersonate it or to build a service that could be confused with its own, and apply to all of its Fear and Greed data, not only the API. CoinMarketCap's two endpoints work on its keyless public API and on every plan of the CoinMarketCap API, including the free Basic tier, at one credit per request.

The inputs are where money appears, and where a reproduction stops being possible:

• Free, from any price feed: bitcoin's realised volatility and drawdowns, momentum, and a volume figure — with the caveats on volume above.
• Free, from a market-cap API: bitcoin dominance and a stablecoin supply ratio, once you choose the coin list in the denominator.
• Paid: social series with published definitions. LunarCrush starts its API at 90 USD a month; Santiment's free plan has 1,000 API calls a month on data 30 days behind, and Sanbase Max at 249 USD a month is its cheapest plan with no lag.
• Paid and licensed: implied volatility from Volmex, or per-instrument option data from which a put/call ratio can be summed — Laevitas issues its API key on a 500 USD a month seat, or charges 0.10 USDC a call without one.
• Not for sale at all: alternative.me's hashtag counts, its paused surveys, its choice of Google queries; CoinMarketCap's site engagement data and its weights.

What you can do about it

Read the components directly, from a source whose definition you can cite. Each input to either index exists as a series of its own, and every one of them is easier to explain than the composite:

• Volatility and momentum: compute them from bitcoin candles — OHLCV from one venue or one aggregator, on a fixed UTC day, with your own window (pull OHLCV from an exchange).
• Volume: the same candles, labelled with their source, after reading why exchange volume numbers disagree.
• Dominance and the stablecoin supply ratio: market capitalisations from the CoinMarketCap API, with the coin list stated.
• Social activity: LunarCrush for mentions, engagement, creators and an engagement-weighted sentiment, or Santiment for document-counted social volume per platform. LunarCrush's own composite, Galaxy Score, names its inputs and not its weights, so it has the same limitation as the indices on this page.
• Implied volatility and options positioning: Laevitas for per-instrument implied volatility and option data (get an implied volatility surface).

For the derivatives market, read the numbers that are not in either index. Neither published methodology lists funding rates or open interest, though both are often described as sentiment. They are measured directly by the venues, and CoinGlass sells them as candles per pair and aggregated across exchanges on every plan of its API (get funding and open interest). A long/short ratio is the derivatives number most often read as a mood, and it counts something narrower than its name; what a long/short ratio actually counts explains what.

Use the index as a dated label, with its publisher's name. What a Fear and Greed value is good for is a short, citable description of conditions on a date — "alternative.me Fear and Greed 48, Neutral, 27 October 2026" — in a report, a chart annotation or a research note. Name the publisher every time: "the Fear and Greed index" with no publisher is two or three different numbers. Never put two publishers' values in one series.

Do not treat it as a forecast, or as a measurement. It is a composite of other numbers, normalised against a lookback, under weights one of the two main publishers does not disclose and inputs neither fully discloses. It says how the inputs compare with their own recent past. It does not say what happens next, and nothing in either methodology claims a test a reader could check.

Get the history by API, and store it as you received it. Both publishers return daily values without a key:

# Alternative.me: every daily value since 1 February 2018, JSON or CSV
curl -s "https://api.alternative.me/fng/?limit=0"
curl -s "https://api.alternative.me/fng/?limit=0&format=csv&date_format=world"

# CoinMarketCap: latest value (updated every 15 minutes) and daily history, 500 per page
curl -s "https://pro-api.coinmarketcap.com/public-api/v3/fear-and-greed/latest"
curl -s "https://pro-api.coinmarketcap.com/public-api/v3/fear-and-greed/historical?start=1&limit=500"
curl -s "https://pro-api.coinmarketcap.com/public-api/v3/fear-and-greed/historical?start=501&limit=500"

Three things to know before you rely on what comes back, each from our own calls on 27 October 2026. Alternative.me's limit=0 returned 3,187 values and four missing days, so check for gaps rather than assuming one row per day. CoinMarketCap's historical endpoint rejected a limit above 500, so page through with start; its reference documents the timestamps as ISO 8601 strings, while the keyless endpoint returned Unix seconds as strings, so parse both. And CoinMarketCap's latest value moves through the day while its history keeps only the 00:00 UTC value, so a figure you read in the afternoon will not be the one that appears in the history for that date.

If the composite is what you need, store the version with it. Neither publisher dates its methodology or keeps a public change log. Record the date you read the methodology alongside the values, and expect that a change to an input — a social platform repricing its API, a source being dropped, the surveys coming back — reaches the series as a change in the number, not as a note.