What do algorithms actually do in markets?

Several different jobs that get lumped together under one word. Market makers quote both sides continuously and earn the spread across enormous numbers of small trades. Statistical arbitrage systems hunt tiny, short lived relationships across many instruments at once. Execution algorithms break large institutional orders into pieces to avoid moving the price. And latency sensitive systems react to new information in fractions of a second. Very little of this is prediction in the way retail traders imagine. Most of it is doing something simple, extremely fast, at very large scale, with costs that only make sense at that scale.

Which edges have algorithms taken completely?

Be honest about this list, because pretending otherwise wastes years. Speed of reaction is gone, since co-located systems on direct feeds act in fractions of a second. Spread capture is gone, because that is the market making business. Simple statistical relationships across liquid instruments are largely arbitraged away. Execution quality on large orders is a solved engineering problem. And processing volume, meaning reading everything published and reacting to all of it, is not something a person can compete with. If your plan depends on being faster or on spotting a simple recurring pattern in liquid markets, that plan is competing directly with the thing machines are best at.

So is there any edge left for a human?

Yes, but it cannot be a slower version of what machines do. It has to be something they structurally cannot do or actively do not want to do. That distinction is the whole answer. The remaining human edges are mostly not skills at all, they are constraints that institutions have and you do not: you can wait indefinitely, you can be completely flat, you can trade in size too small to be worth anyone's infrastructure, and you can be selective to a degree no professional operation can justify. Those are structural rather than clever, which is why they are durable.

Where Is the Edge in Trading in an Algorithmic World?

Q: What is the biggest structural advantage a retail trader has?

Time horizon, and it is not close. Professional managers face redemption pressure, monthly reporting, benchmark comparison and career risk. They frequently cannot hold a position through a drawdown that will resolve in six months, and they cannot sit in cash for a quarter while nothing good is available, because clients ask what they are paying for. You can do both. Nobody redeems from you, nobody compares you to an index in March, and cash is a legitimate position rather than an admission of failure. That freedom is worth more than any indicator and it costs nothing to use.

Q: Does small size help or hurt?

Both, and the helpful part is underrated. Small size means you can enter and exit without moving the price against yourself, which is a genuine problem for anyone running real money. It also means you can trade situations too small to be worth an institution's attention, since a strategy that can only absorb a modest amount of capital is not worth building infrastructure around. Capacity constraints are real and they protect small participants. The unhelpful part is that fixed costs weigh more heavily on a small account and one bad trade is a larger share of it, so the advantage only survives if the risk control does.

Q: Can algorithms handle a change in market regime?

Less well than they handle everything else, and this is where judgment earns its keep. Systems are fitted on history, so they are strongest when the future resembles the past and weakest when the question the market is asking changes. When the market stops trading inflation and starts trading growth, the relationships a model learned stop describing the world, and it takes time and data before that shows up in the fitting. A human who understands why the market cared about something can notice the shift from the behaviour rather than waiting for the statistics to confirm it.

Q: What is not an edge, however hard you work?

Five things worth crossing off. Reacting faster, because you cannot. Indicators on a chart, since they are computed from the same price everyone has and involve no information anyone else lacks. Chart patterns by themselves, for the same reason. Reading order flow faster than machines that see the same book in microseconds. And simply working harder, because more screen time mostly produces more trades rather than better ones. None of these are worthless as tools. They are just not edges, because an edge has to be something not equally available to everyone with the same screen.

Q: Does AI close all of this?

It closes some of it and leaves the important part alone. Machine learning is very good at extracting patterns and processing far more material than a person can read, and it will compete away more of the simple, repeatable retail approaches over time. What it does not change is the structural advantages, because those are about constraints rather than capability. An AI managing institutional money still faces redemptions, still reports monthly, still needs capacity and still cannot sit in cash for a quarter. Crowding into similar models can even improve conditions for patient capital, because when many participants are positioned the same way, the unwind is the opportunity.

Q: How do I work out which edge is actually mine?

Ask what you can do that a well funded professional operation cannot or will not. Can you wait longer than they can, because nobody is judging you quarterly? Can you trade something too small for them to bother with? Can you sit out for weeks when nothing is there? Can you understand why a market is moving rather than only that it moved? Those are real answers. If your honest answer is that you look at charts harder or react quickly, that is not an edge and no amount of effort converts it into one. The uncomfortable version of this question is the useful one.

Q: What does this look like in practice?

Trade less and choose better, since selectivity is free to you and expensive to anyone with infrastructure to justify. Let the machines have the first move on a release and act on the one that holds, because being late to a real move costs a few ticks while being early to a false one costs the trade. Hold a view for as long as it takes rather than as long as a reporting period allows. Understand the reason behind a move rather than only the shape of it. And accept being flat, because the ability to do nothing is a genuine advantage that most people give away for free.

Helious Desk · 8 August 2026 · 7 min read

Machines have permanently taken several things that used to be edges, and they are not giving them back. The honest answer is neither despairing nor comforting. What they left behind is real and it lasts, but it is made of constraints rather than skill, which is not where most people think to look.

1. What do algorithms actually do?

Several different jobs get flattened into one word. Market makers quote both sides all day and earn the spread across a very large number of small trades. Statistical arbitrage hunts tiny relationships that only hold for a short while, across many instruments at once. Execution algorithms have a duller job, slicing a large institutional order into pieces so it does not move the price. Then there is the latency sensitive work, which acts on new information in a fraction of a second.

Prediction barely appears on that list, at least not the kind retail traders picture. Most of this is something simple, done extremely fast, on a cost base that only works at very large scale. The cost base is the part that matters.

2. Which edges have they taken completely?

Worth being blunt about this, because pretending otherwise costs people years.

Speed of reaction is gone. Co-located machines sit in the exchange's own building, read the direct feed and act in fractions of a second, as we covered in does smart money know the number before it prints. Spread capture is gone too, because that is the market making business. Simple statistical relationships across liquid instruments have mostly been arbitraged away, traded until there is nothing left in them. Execution quality on large orders is a solved engineering problem. And processing volume, reading everything that gets published and reacting to all of it, was never a human contest.

If your plan needs you to be faster, or to spot a simple repeating pattern in a liquid market, you are competing head on with what machines do best. It is not close, and working harder does not change it.

3. So is there any edge left?

Yes, but it cannot be a slower version of what machines do. It has to be something they cannot do at all, or something they do not want to do.

What survives is mostly not skill. It is a set of constraints that institutions have and you do not. You can wait as long as you like, and you can be flat, holding nothing at all. You can trade in size too small to be worth anyone's infrastructure, and you can be pickier than any professional operation could justify being.

None of that is clever. It is how you are built, and that is why it lasts. Nobody competes it away, because doing so would mean giving up on being a large institution.

4. What is the single biggest advantage?

Time horizon, and it is not close.

Professional managers live with redemptions, which means clients pulling their money out. They file monthly reports, they are measured against a benchmark index, and they have careers to protect. So holding through a drawdown that will resolve in six months is frequently not an option. Nor is sitting in cash for a quarter while nothing good is available, because clients ask what they are paying for.

You can do both. Nobody redeems from you, nobody compares you to an index in March, and cash is a legitimate position rather than an admission of failure.

That freedom is worth more than any indicator and it costs nothing. Most retail traders give it away because they feel obliged to be in the market, and nobody is making them. You put that pressure on yourself, so you can take it off.

5. Does small size help or hurt?

Both, and the helpful half is badly underrated.

What small size buys you

You can get in and out without pushing the price against yourself. For anyone running real money that is a genuine problem and an expensive one. You can also trade situations too small to be worth an institution's attention, because a strategy that only absorbs a modest amount of capital is not worth building infrastructure around. Capacity constraints are real, and they protect small participants.

What it costs you

Fixed costs weigh more heavily on a small account, and one bad trade is a bigger share of it. The advantage only survives if the risk control does.

6. Can algorithms handle a regime change?

Less well than they handle everything else. This is where judgment earns its keep.

A system is fitted on history, so it is strongest when the future looks like the past and weakest when the question the market is asking changes. When the market stops trading inflation and starts trading growth, the relationships a model learned stop describing the world, and it takes time and fresh data before the fit catches up.

A person who understands why the market cared about something can see the shift in how it behaves, instead of waiting for the statistics to confirm it. That is the argument in why market narrative is what your trading is missing, and it is the most defensible human edge here. The first visible sign is usually a familiar correlation breaking down.

7. What is not an edge, however hard you work?

You cannot react faster, so cross that off first. Indicators on a chart are worked out from the same price everyone else has, so they carry no information anyone else lacks, and chart patterns on their own fail for the same reason. You will not read order flow faster than machines that see the same order book in microseconds either, as what happens to the DOM during a news release shows. And working harder is not an edge at all. More screen time mostly produces more trades, not better ones.

None of these are worthless as tools. They are just not edges. An edge has to be something the person at the next screen does not have.

8. Does AI close all of this?

It closes some of it and leaves the important part untouched.

Machine learning is very good at pulling patterns out of data, and it reads far more material than a person ever will. Over time it will compete away more of the simple, repeatable retail approaches. If your method is a mechanical rule applied to a liquid market, expect that.

What it does not change is the structural advantages, because those are about constraints rather than capability. An AI running institutional money still faces redemptions, still reports monthly, still needs capacity, and still cannot sit in cash for a quarter. A better model does not rewrite a mandate.

Crowding has a second order effect as well. When many participants pile into similar models they end up positioned the same way, and the unwind of a crowded position is an opportunity for whoever was patient enough to stay out of it.

9. How do I find which edge is mine?

Ask one uncomfortable question. What can you do that a well funded professional operation cannot or will not?

Can you wait longer, because nobody judges you quarterly? Can you trade something too small for them to bother with, or sit out for weeks when nothing is there? Can you understand why a market is moving, rather than only that it moved?

Those are real answers. If your honest answer is that you stare at charts harder, or that you react quickly, that is not an edge, and no amount of effort turns it into one. Facing that early is what shortens the learning curve we described in how long it takes to become profitable.

10. What does this look like in practice?

Trade less and choose better, because being picky is free to you and expensive to anyone with infrastructure to justify. Let the machines have the first move and act on the one that holds. Being late to a real move costs a few ticks. Being early to a false one costs the trade.

Hold a view as long as it takes, not as long as a reporting period allows, and work out the reason behind a move rather than only the shape of it. Then accept being flat. Doing nothing is a genuine advantage, and most people give it away for free.

Where the terminal fits

Follow that logic and the tooling question answers itself. You are not trying to out-run anyone, so none of this is about speed for its own sake. The job is to understand what happened while it still matters, which is a different job and a reachable one.

That means knowing what is scheduled, through the calendar and alerts. Once a release lands you want a surprise score telling you whether it was a genuine shock rather than a guess from the headline, you want to know whether the curve agreed, and you want one plain read on why the tape is where it is. All of that is context fast enough to act on rather than a speed race you were never going to win. It is $39.99 a month with a free tier, and the methodology page shows the workings.

Where to go next

The speed question gets a whole post to itself in does smart money know the number before it prints. The most defensible human edge is the one covered in market narrative. If you want realistic timelines rather than encouragement, read how long it takes to become profitable, and for the early signs of a regime shift there is correlations and market news. What is coming next sits on the economic calendar and the data hub.

Helious is built for the edge you can keep, and that edge is not speed. It puts the calendar, every scored release, the curve and one plain read on the tape in one place, so you can understand what just happened while it still matters. $39.99 a month with a free tier. Built by traders, for traders.

This post is general information and not financial advice. Nothing here suggests that any approach produces profits, the majority of retail traders lose money, and trading involves substantial risk.

Launch the terminal →