Crypto Fear and Greed Index: methodology and limits

The crypto Fear and Greed Index is a provider-defined sentiment score, usually on a 0–100 scale, that combines selected market and attention measures. Lower readings represent fear and higher readings represent greed. Its meaning depends on the publisher, ingredients and update time; a reading is neither a probability of a price rise nor a complete assessment of an individual token.

Before interpreting a number, identify whose index it is. Alternative.me and CoinMarketCap publish distinct series under similar names. Their public methods, reviewed on 27 October 2026, describe different coverage and inputs. The sections below compare those methods and work through a hypothetical score change.

Start with the publisher and the scale

Treat a score as a summary in the publisher's own units. A hypothetical reading of 70 does not establish that 70% of investors are bullish, that a coin has a 70% chance of rising, or that it is 70% overvalued. Those interpretations would require separate definitions and evidence.

Likewise, a rise from 30 to 50 is a change of 20 index points. It does not imply a 20% expected return. Even the apparent precision of an integer can conceal transformations, rounding and several inputs moving in opposite directions.

Keep the provider's classification beside its value. Avoid imposing a zone table copied from another website unless it belongs to that exact series and method. A label such as extreme fear describes a bucket; it supplies no exit date, price target or assurance that conditions will improve.

What Alternative.me publishes

Alternative.me's methodology section explicitly says its current index covers Bitcoin. Its listed ingredients are volatility and drawdowns at 25%, momentum and volume at 25%, social activity at 15%, surveys at 15%, Bitcoin dominance at 10%, and Google Trends at 10%. The survey component is marked paused. Volatility and momentum inputs use comparisons with 30-day and 90-day averages. Alternative.me's published methodology.

The non-survey weights add to 85%. That arithmetic does not reveal how the production calculation handles the pause. The published explanation does not provide a complete formula mapping raw inputs to the score or explain the effective redistribution. Renormalizing the remaining weights yourself would produce an assumed model, rather than a verified reconstruction.

This distinction matters when a tutorial promises to calculate today's exact index from public prices. Matching one historical value does not establish that its data sources, transformations and treatment of missing components match the provider. Ask what evidence would make the reconstruction reproducible across many dates.

CoinMarketCap uses a different set of inputs

CoinMarketCap describes its CMC index as proprietary. Its current chart page lists five families: momentum among the ten largest cryptocurrencies by market capitalization excluding stablecoins; Bitcoin and Ethereum implied volatility through BVIV and EVIV; BTC and ETH options put/call ratios; the Stablecoin Supply Ratio, comparing Bitcoin's capitalization with that of major stablecoins; and CMC search and engagement data. CMC's current index methodology.

CMC's original introduction also describes a broader market approach. Broader coverage does not mean an equal survey of every token holder. The inputs still have specific asset and platform boundaries.

The current explanation does not disclose every numeric weight and transformation needed to independently reproduce the series. Consequently, an Alternative.me value and a CMC value should retain separate names in a journal or dataset. A disagreement is a prompt to inspect measurement differences. Averaging them would create another composite whose interpretation you would need to define.

A checked example: one score, different stories

The following teaching model is entirely hypothetical. It reproduces neither provider's calculation. Its component scores are invented, already normalized to 0–100, and oriented so that a higher score raises the composite.

Define three components: price strength with a 40% weight, calmness with a 40% weight, and attention with a 20% weight. Calmness here is an invented inverse-risk score; it is not a raw volatility reading. The weights total 100%.

The starting score is 0.40 × 30 + 0.40 × 25 + 0.20 × 50 = 32. In A, strength and calmness each add ten points relative to the start, producing a 20-point increase. In B, calmness and attention each add ten points instead. Both later scores equal 52, despite different underlying changes.

These are alternative scenarios, not successive days. The example demonstrates information loss when several measurements become one number. The headline change cannot identify its cause.

There is another reason to inspect the ingredients. If price strength and calmness both respond to one price recovery, two rising components need not be independent confirmations. Adding the composite beside momentum in a model can partly reuse the same underlying information. Whether the index contributes anything useful must be tested against the inputs already available.

Check what the score leaves unanswered

A sentiment composite cannot replace the measurements required for a particular decision. Consider three boundaries.

First, expected movement and price direction are different questions. Volmex describes BVIV and EVIV as option-derived measures of expected volatility over a constant 30-day horizon. That is distinct from measuring past price variation, and an expectation is not a guaranteed future outcome. Volmex's index documentation. If your rule needs a realized-volatility gate, define it separately; the volatility regime filter worksheet shows why the measurement and resume rule need their own specification.

Second, attention needs interpretation. Google explains that Trends uses sampled search requests normalized by time and geography, with relative interest scaled to 0–100. It is not scientific polling. Google's explanation of Trends data. A search spike might reflect curiosity after bad news. The score alone does not establish a positive attitude or a purchase.

Third, an aggregate mood says little about a token's ownership, supply schedule or executable market depth. The distinction between active addresses and actual users prevents activity from becoming an invented headcount. The market cap versus fully diluted valuation guide supplies a separate supply-definition check. Neither task is completed by a fear or greed label.

Keep the observation clock

A stored daily value needs an availability record before it can support a historical decision. Alternative.me's documented API response includes a value, classification and timestamp, with time_until_update for the latest value. Those fields help identify an observation; they do not substitute for your own receipt log. Alternative.me's API documentation.

For a hypothetical timing example, suppose a value carries an 08:00 UTC observation timestamp, is published at 08:05 and reaches your system at 08:05:03. A decision at 08:02 cannot use it. These times are invented and do not describe either provider's release schedule.

Joining the value to every trade bearing the same calendar date would hide that error. Preserve the provider timestamp, observed publication time when available, local receipt time and decision time. When publication history is absent, distinguish an assumed delay from an observed one. The look-ahead bias audit explains how to test whether later information changes earlier decisions.

Repeated downloads of the same observation are not fresh sentiment events. If a feed stops updating, retaining its last value may make a dashboard look calm while its evidence grows stale. Specify how an overdue observation is identified and whether a dependent rule pauses, requests review or remains in a declared fallback state.

Use a reading with a small evidence worksheet

A useful journal entry should let another reader reconstruct what the number meant at the time. This worksheet can be copied into a research log.

For discretionary review, the intended role might simply be to record context before examining a trade. For a systematic experiment, write the exact rule before evaluating outcomes. A threshold discovered after inspecting a large price rebound has already benefited from that outcome.

Useful failure checks follow directly from the worksheet. Flag a provider change that retains the same column name. Investigate a stale timestamp even if the endpoint responds successfully. Reject an explanation that attributes a score jump to social activity without component evidence. Keep an unexplained score change unexplained.

Test added value before turning context into a trigger

A claim that extreme fear predicts an attractive entry needs more than examples of rebounds. Define the instrument, decision delay, holding period, exit rule, costs and comparison strategy. Freeze the provider and threshold selection before evaluating a later period. Record every alternative tried, including versions that did poorly.

Compare the rule with and without the index while keeping the remaining procedure fixed. Inspect exposure time, number of decisions, turnover and losses alongside returns. If a filter simply leaves capital uninvested for longer, report that change instead of assigning the whole difference to superior timing.

This article has not performed that predictive test. Its numerical example verifies the mechanics of an invented composite, and its timing example checks when a reading becomes usable. For the next index value you encounter, save its provider, timestamp and intended role first. Those three fields turn an isolated mood label into a piece of evidence you can assess.