In Crypto, the Best Trading Algorithms Should Know When to Do Nothing

Crypto never closes, but opportunity is not constant. As AI moves deeper into trading, the next competitive edge may be knowing when a valid signal does not deserve capital.

Crypto’s 24/7 market has encouraged a simple assumption: if a machine can trade continuously, it should.

After roughly 15 years of trading futures, I have come to believe almost the opposite. Some of the most valuable trading decisions are not entries or exits. They are the moments when the best position is no position at all.

That lesson became even clearer when I moved from discretionary trading toward automated trading technology.

Turning a recognizable setup into software is one challenge. Teaching a machine when that same setup should be ignored is much harder.

An experienced trader can look at a familiar market and think:

Not today.

A machine needs a reason.

As artificial intelligence becomes more deeply integrated into crypto trading, that distinction may matter more than the industry’s obsession with producing better predictions.

Crypto is always open. Opportunity is not.

Bitcoin, Ether and other crypto assets trade around the clock, across weekends and multiple venues.

Perpetual futures add continuous leverage and funding-driven positioning. For an algorithm, there is almost always something happening and therefore almost always a reason to act.

But activity is not opportunity.

Liquidity can change sharply between a busy weekday and a quiet weekend. Funding can become crowded. A seemingly orderly market can suddenly turn into a liquidation-driven move.

The signal may look familiar while the market around it has changed.

Experienced traders learn to recognize these moments. A setup can be technically valid and still be a poor trade because the conditions that once made it profitable are no longer present.

This is difficult to translate into code.

A discretionary trader may notice a change in volume, speed, liquidity or positioning and simply conclude that the market feels different.

Software cannot rely on that instinct. It needs to know what changed, whether it matters and whether the strategy’s edge is still there.

A mature trading system, then, is not simply a collection of entry signals.

It is also a collection of reasons not to trade.

AI could become very good at finding reasons to trade

Much of the enthusiasm around AI in trading centers on prediction.

Can a model anticipate Bitcoin’s next move? Can it combine price, volume, onchain data, derivatives positioning, sentiment and news faster than a human trader?

Probably.

But that solves only part of the problem.

Crypto generates an endless stream of information. AI can generate an equally endless stream of interpretations.

The danger is confusing an interpretation with an opportunity.

A model may detect bullish sentiment, rising volume or changing derivatives positioning and still be looking at a market where the risk-reward is poor.

The better question is not simply:

“What does the model think happens next?”

It is:

“Are the conditions in which this model deserves to be trusted actually present?”

That question is harder because it requires a system to recognize the limits of its own confidence.

One of the most important outputs of an advanced trading system should therefore be neither “buy” nor “sell.”

Sometimes it should be:

I don’t know.

In trading, uncertainty is not necessarily weakness. Recognizing when not to deploy capital can be a form of intelligence.

Doing nothing can be a successful decision

There is an obvious objection.

Market makers, arbitrage systems and high-frequency strategies are designed to stay active. They cannot wait for a handful of perfect directional setups.

But even these systems have limits.

A market maker may reduce exposure when volatility rises too quickly. An arbitrage system may detect a price discrepancy but conclude that execution risk makes it untradeable.

The principle remains the same:

Activity should never be confused with effectiveness.

The difficulty is that trades are easy to measure. They have entries, exits and outcomes.

A rejected trade leaves almost nothing behind.

Yet that missing trade may represent one of the best decisions the system made.

Abstention should therefore be treated as part of a strategy’s intelligence, not as a failure to find an opportunity.

The next edge may be knowing when to stop

As AI models, computing power and market data become more accessible, generating another trading signal will become less impressive.

More firms will use sophisticated models. More traders will analyze onchain activity, derivatives data and sentiment in real time.

If prediction becomes easier to obtain, the competitive advantage may move elsewhere.

The real edge may lie in knowing when an attractive-looking prediction should not be trusted.

Can a model recognize that current market conditions no longer resemble the environment in which it performed well?

Can it distinguish a genuine opportunity from a familiar pattern appearing during a liquidity vacuum or a liquidation-driven move?

Can it reduce exposure when uncertainty rises instead of producing another confident signal?

Those questions may ultimately matter more than squeezing another percentage point of accuracy from a prediction model.

For years, one of the strongest arguments for algorithmic trading has been that machines do not hesitate.

That advantage is real. Emotion-driven hesitation can ruin a good trading decision.

But hesitation and selectivity are not the same thing.

Removing emotion should not mean removing doubt.

A market being open is not a reason to trade. A signal being present is not a reason to execute. And an AI model being capable of producing a prediction does not mean that prediction deserves capital.

After years of trading futures and later working on translating trading logic into automated systems, I have become increasingly convinced that the hardest part of automation is not teaching machines how to act.

It is teaching them when not to.

In crypto, where markets never sleep and leverage can punish mistakes quickly, that ability may become increasingly valuable.

The future of algorithmic trading may belong not to the system that finds the most opportunities, but to the one that rejects the most bad trades.

The most intelligent trading machine may ultimately be the one that knows when it doesn’t know.

---

Amir Naser Hojati is the founder of fintech company and a futures trader with 15 years of experience trading international futures, including CME markets. His work focuses on translating discretionary market analysis and volume-based trading concepts into automated trading technology.