How to Read Crypto Market Signals Without Fooling Yourself

What counts as a market signal

A market signal is any observable measurement a trader uses to form a view about price. That definition is deliberately wide, because the practical problem is not finding signals. It is that every dashboard offers hundreds of them, they are presented with equal visual authority, and almost none of them come with an honest statement of what they can and cannot do.

This article is about interpretation rather than causation. If you want the underlying drivers of price itself, read what actually moves crypto prices first, which covers liquidity conditions, supply events, and flows. This one starts a layer above that: given the measurements traders actually stare at, which are informative, and about what.

Three properties separate a useful signal from a decorative one.

Timing relative to the move. A leading signal changes before price. A coincident signal changes at the same time as price. A lagging signal changes after price. Almost every popular crypto indicator is coincident or lagging, and the marketing language around them systematically implies otherwise.

What it is actually measuring. Many indicators are transformations of the same input. A moving average, a momentum oscillator, and a trend line all take price as their only input. They are three views of one variable, not three independent pieces of evidence.

Whether it can be wrong. A signal you can only interpret after the fact is not a signal, it is a narrative. If there is no reading that would have told you to do the opposite, the indicator is not carrying information.

The three families of crypto signals

Nearly everything on a trader's screen belongs to one of three families, and the families differ in what kind of error they are prone to.

The value of this split is that it tells you when you are double-counting. If your bullish case rests on a moving average crossover, a bullish engulfing candle, and an oscillator turning up, you have one signal counted three times, because all three are functions of the same price series. Adding a structural signal such as open interest, or a sentiment signal such as a positioning survey, genuinely adds a second dimension. Adding a fourth price indicator does not.

Family one: price and technical signals

This is the largest family and the one with the weakest average evidence base, for a structural reason: it is the family where it is easiest to search for a rule that worked on past data.

Suppose you test moving-average crossover rules with every combination of fast and slow lengths between 5 and 200 periods. That is close to twenty thousand rules. If none of them had any genuine edge, you would still expect roughly a thousand of them to look profitable at a 5% significance level purely by chance. The one that looks best in the backtest is, more often than not, the luckiest rather than the truest.

What this actually tells you

A backtest result presented without its search space is uninterpretable. The relevant question is never "did this rule work" but "how many rules were tried before this one was shown to me." A single rule tested once and a single rule selected from twenty thousand look identical on the chart and mean completely different things.

This is why simpler is usually more honest here. Rules with few free parameters have less room to be fitted to noise. A plain trend filter with one lookback has a smaller search space than an indicator with five tunable inputs, so a given level of past performance is more likely to be real.

The actionable version. When you adopt a technical rule, write down the parameters before you test them, and test the neighbouring parameter values too. If a 14-period setting works and 12 and 16 do not, you have found an artefact. If the whole neighbourhood behaves similarly, you have at least found something stable.

The individual tools in this family are covered separately in how to read a crypto candlestick chart and what the RSI indicator can and cannot tell you. Both articles include the reliability caveats that most tutorials omit.

Family two: sentiment signals

Sentiment signals attempt to measure what the crowd feels or how it is positioned. The best-known one in crypto is the Crypto Fear and Greed Index, published by Alternative.me, which combines volatility, market momentum and volume, social media activity, Bitcoin dominance, and Google Trends data into a single number from 0 to 100. A survey component is also published in the methodology but is currently marked as paused.

That composition is the honest answer to why the index is coincident. It cannot lead price by much, because most of its inputs are price. The full construction, and the contrarian strategy people build on it, are examined in what the Fear and Greed Index actually measures.

Sentiment signals also suffer from reflexivity. If a widely watched gauge prints extreme fear and a large number of traders buy that reading, the gauge has changed the thing it was measuring. Any sentiment measure popular enough to be tradable is popular enough to be self-affecting.

What sentiment signals are genuinely good for

They are poor timing tools and reasonable context tools. Knowing that positioning is crowded on one side tells you something real about the shape of the risk you are taking: crowded positioning means a shock is more likely to produce a disorderly move, because more participants need to exit through the same door. That is a statement about the distribution of outcomes, not about direction, and it should change your position size rather than your position.

For the psychology behind why crowds cluster, crypto market psychology covers the emotional cycle in detail.

Family three: structural and on-chain signals

This family measures the market's plumbing rather than its mood: how much leverage is outstanding, who is paying whom to hold it, how deep the order book is, and how capital is distributed across assets.

Every row shares the same defect: the reading is real, but its sign is ambiguous without additional context. This is the family's characteristic failure mode, and it is different from the technical family's. Technical signals fail by being fitted to noise. Structural signals fail by being genuine measurements whose direction of implication depends on facts the measurement itself does not contain.

What this actually tells you

Structural signals are most useful as conditional statements, not standalone ones. "Open interest is high" is not actionable. "Open interest is high and funding is strongly positive and price has stalled" describes a specific, identifiable configuration in which a downward move has more available fuel than usual. The conjunction carries the information; no single term does.

The occasional genuinely leading signal lives here. Positioning data can lead, because leverage must be built before it can be unwound. That does not make it a forecast of direction, but it does tell you where the exits are crowded before the crowding matters.

The practical use is risk sizing. When these signals line up, the correct response is usually to reduce size or widen stops rather than to flip direction. You are being told the distribution has fatter tails, not which tail.

The individual mechanics are covered in open interest in crypto, how funding rates work, how to read a crypto order book, and what Bitcoin dominance measures.

The limitation every signal shares

Strip away the differences and all three families share one constraint: crypto markets are close enough to efficient at the level of publicly available indicators that a widely known signal with a large, stable edge should not survive.

This is not a claim that crypto markets are perfectly efficient. They plainly are not. It is a narrower and more defensible claim: any signal that is (a) publicly documented, (b) computable from free data, and (c) visible on every major charting platform has been examined by an enormous number of well-capitalised participants. If it carried a large and reliable edge, that edge would be traded away until it did not.

What survives that filter tends to be one of three things: effects that are hard to capture because of costs and capacity limits, effects that only work in specific regimes and lose money in others, or effects that require accepting genuine risk rather than exploiting a mistake.

What the evidence actually supports

Peer-reviewed research on crypto returns is much thinner than the volume of indicator content implies, but it is not empty, and it points in a specific direction.

Liu and Tsyvinski, in Risks and Returns of Cryptocurrency, establish that crypto returns are not explained by standard stock market or macroeconomic factors, and identify two crypto-specific effects: a strong time-series momentum effect, and proxies for investor attention that forecast returns.

Liu, Tsyvinski and Wu, in Common Risk Factors in Cryptocurrency, construct crypto equivalents of a wide range of factors studied in equities and find that a three-factor model of market, size, and momentum accounts for the returns of the long-short strategies they test.

What this actually tells you

Momentum has the strongest published support, and it is not a chart pattern. The momentum effect in this literature is a systematic, cross-asset, rules-based construction with defined lookbacks and rebalancing, not a discretionary read of whether a chart "looks strong." The evidence supports the former and says nothing about the latter.

Attention is measurable and matters, which is unusual. Search interest and social volume forecasting returns is a genuinely non-obvious finding, and it is the strongest academic case for taking any sentiment-adjacent measure seriously. Note carefully what it does not say: it does not say that a sentiment index composed largely of price data forecasts returns.

Almost nothing else in the retail indicator canon appears. Candlestick patterns, oscillator thresholds, dominance-based rotation rules, and max pain do not have comparable published support. That is not proof they are worthless, but it does mean the burden of evidence sits with the person claiming they work, and you should treat them as heuristics rather than as findings.

The actionable version. If you want an evidence-aligned starting point, a mechanical trend or momentum rule applied consistently across assets has more support behind it than any discretionary pattern read, and it has the additional virtue of being falsifiable. Everything else on your screen should be treated as context that shapes position size, not as a trigger.

How a signal with a good hit rate still misleads you

This is the arithmetic that explains why so many indicators feel useful and perform badly, and it is worth working through slowly.

Take a signal with genuinely impressive credentials. It fired before 8 of the last 10 major drawdowns. That is an 80% hit rate on the events that matter, and it is the statistic that gets advertised.

Now add the two numbers that never get advertised: how often the signal fires when no drawdown follows, and how often drawdowns happen at all. Say the signal also fires in 20% of calm periods, and major drawdowns occur in 10% of months.

Over 100 months:

The signal fires 26 times. It is right 8 of those times. Its actual accuracy when it speaks is 8 divided by 26, about 31%.

The table has the numbers; the grid has the intuition.

Reading Figure 1

Count the coloured dots, not the grey ones. The signal spoke 26 times, and those are the only months where it asked you to do anything. Twenty-six dots out of a hundred is the entire surface on which this indicator can help or hurt you.

Now compare the two coloured groups. Eight green against eighteen red. Every time this signal warns you, the red group is the more likely explanation of why it is speaking — better than two to one. That ratio is the 31% figure, and it is visible without doing any arithmetic.

The two lonely dots matter more than they look. Those are the drawdowns that arrived with no warning at all. They are the reason a signal cannot be used as an all-clear: staying invested because the indicator is quiet is a bet that you are in the grey block, and 2 of the 74 quiet months were not.

And notice what changes the picture. The red group is large not because the signal is bad but because calm months outnumber drawdowns nine to one. Improve the signal's hit rate from 8 to 10 out of 10 and you add two green dots. Halve its false-positive rate and you remove nine red ones. The second is worth four times the first, which is the opposite of where indicator marketing puts its attention.

What this actually tells you

The 80% figure and the 31% figure are both true, and they answer different questions. Eighty percent answers "given a drawdown, did the signal warn me." Thirty-one percent answers "given a warning, should I expect a drawdown." Only the second one is a question you can trade. Indicator marketing consistently reports the first.

The base rate does most of the work. The reason the number collapses is that calm months outnumber drawdown months nine to one, so even a modest 20% false-positive rate generates more than twice as many false alarms as true ones. This is not a flaw in the signal. It is what happens to any test for a rare event, and it is why medical screening tests for rare diseases have the same problem.

Acting on every firing is worse than useless once costs are included. Twenty-six exits over 100 months, of which 18 were unnecessary, means 18 round trips of fees, spread, and re-entry at a worse price, plus the times you were out during a recovery. The signal can be genuinely informative and still lose money net of execution.

The actionable version. Before adopting any signal, demand three numbers rather than one: how often it fires before the event, how often it fires without the event, and how often the event happens. If the person promoting it cannot supply the second and third, they have not measured the thing they are claiming. And when the precision is low but not negligible, the correct use is to reduce size on a firing, not to exit entirely, because a 31% signal is real information and terrible as a binary switch.

How to combine signals without confirmation bias

Confirmation bias in trading is not usually a failure of intelligence. It is a failure of sequence. Most traders form a view first, then consult indicators, and the enormous menu available guarantees that some subset will agree with whatever they already thought.

Three habits break that sequence.

A checklist for reading any new signal

Conclusion

The central fact about crypto market signals is unglamorous: nearly all of them are measurements of the present, dressed in the language of the future. Candles, oscillators, sentiment indexes, dominance ratios, funding, and open interest are all real quantities, and every one of them is largely a description of what has already occurred. Treating a coincident measurement as a forecast is the single error that produces the most confident losing trades.

The three families fail in different ways, which is exactly why the split is useful. Price-derived signals fail through data mining, because the space of testable rules is effectively infinite. Sentiment signals fail through reflexivity and, in the most popular case, through being substantially built from price data in the first place. Structural signals fail through sign ambiguity, since the same reading is frequently consistent with opposite outcomes. Recognising which failure mode you are exposed to matters more than the specific settings you use.

The published evidence is narrower than the indicator ecosystem suggests. Time-series momentum and investor attention have peer-reviewed support in crypto specifically; a mechanical, rules-based momentum construction has more behind it than any discretionary pattern read. Nearly everything else in the retail canon is heuristic, and should be held to the standard of a heuristic.

Finally, the base-rate arithmetic deserves to be internalised rather than merely read. A signal that catches 80% of drawdowns can be right under a third of the times it fires, because the event is rare and false alarms are not. That is not a reason to discard signals. It is a reason to use most of them to size positions rather than to trigger them, to demand the false-positive rate as routinely as the hit rate, and to write down what would prove you wrong before you go looking for what proves you right.

Frequently asked questions

Sources and further reading

Primary sources:

• Liu and Tsyvinski — Risks and Returns of Cryptocurrency (NBER Working Paper 24877)
• Liu, Tsyvinski and Wu — Common Risk Factors in Cryptocurrency (NBER Working Paper 25882)
• Alternative.me — Crypto Fear and Greed Index methodology

Related CoinBeaver articles:

• What actually moves crypto prices
• How to read a crypto candlestick chart
• What trading volume tells you
• What the Fear and Greed Index actually measures
• What Bitcoin dominance measures
• How to read a crypto order book
• What the RSI indicator can and cannot tell you
• Open interest in crypto
• How funding rates work
• Crypto market psychology

This article is educational and is not financial advice. The 100-month worked example uses illustrative hit rates, false-positive rates, and base rates chosen to demonstrate the arithmetic; they are not measurements of any specific indicator. Signal methodologies and index compositions change, so verify the current construction of any index before relying on it.