The Pitch

A note on the industry that has grown up around retail trading: what is being sold, why a working method is a strange thing to sell, and what actually determines whether an approach makes money. This page criticises a market structure, not any individual. Plenty of honest teachers work in this space.

Start with what an edge is

An edge is a claim that a market is mispricing something in a way that has not yet been competed away. That definition has two consequences most marketing quietly ignores.

The first is that an edge is a relative property. It exists only in relation to what everyone else is doing. It is not a fact about a chart pattern; it is a fact about the current population of participants and their behaviour. Change the population and the edge changes with it, without anyone announcing the change.

The second is that edges are consumed by use. Every participant acting on the same observation pushes price toward the thing they were exploiting. This is not a theory anyone needs to defend abstractly. It is the mechanical consequence of many people placing similar orders at similar times. The more widely an observation is known, the less room remains in it.

An edge, then, is a depleting resource with a population-dependent shelf life. Hold that thought while considering why someone would sell one.

The case of the systematic method

Suppose someone has an algorithm that genuinely makes money. Why would they sell it rather than run it?

There are honest answers, and they are worth stating properly rather than dismissing.

• Capacity. Many real edges are small. A method that produces good returns on a modest account may simply stop working at ten times the size, because the orders required start moving the very prices the method depends on. If the method cannot absorb more capital, the owner cannot scale it by trading it. Selling access is one of the few ways to increase revenue from a capacity-constrained method — and it also destroys the method for everyone including the seller, which is the part rarely mentioned.
• Revenue shape. Trading income is volatile, arrives unpredictably, and comes with drawdowns that have to be survived psychologically as well as financially. Subscription income is smooth, arrives monthly, and has no drawdown. Many people would rationally prefer the second even at a lower expected value. This is a legitimate business decision.
• Diversification. An operation that runs several methods may license one it considers least valuable, keeping the rest.

And there are the less honest answers, which are structurally more common because they cost less to produce.

• The method already decayed. It worked, it was crowded out, and what remains saleable is the historical record of when it worked. The track record is real. The forward expectation is not.
• It never worked outside the sample. A backtest can be made to look excellent by anyone with a computer and enough patience, and the techniques that produce this are not exotic. More on that below.
• The product is the audience, not the method. Where the seller earns rebates from a broker, commissions on volume, or affiliate income from a data vendor, their revenue depends on subscribers trading, not on subscribers profiting. Read the incentive rather than the claim.

Note what the honest answers have in common: they involve capacity limits, lockups, high prices, and a small number of counterparties. That is roughly what licensing to institutions looks like. It is not what a public funnel with a monthly fee and unlimited seats looks like. When a method is sold to an unbounded number of buyers at a price an individual can afford on impulse, the capacity argument has been abandoned, and with it the main honest reason for selling.

The case of the discretionary method

The other product is judgement: someone who genuinely trades well, teaching you to see what they see.

This one fails differently, and more sympathetically, because the seller is often sincere.

Skilled discretionary traders are running on a large body of knowledge they cannot fully articulate. They have watched the same instrument for years and absorbed how it behaves at the open, before a number, into a holiday, in a thin session, when it has already moved twice. Most of that never becomes an explicit rule because it never was one. Ask them to write down the method and they will produce a description that is true, teachable, and radically incomplete, because the parts they can state are the parts that were easy to state.

What is missing is not information you can be given. It is the accumulated experience that tells you which of the stated rules to ignore today, and that comes from having been wrong in specific ways many times.

Then there is everything about the trader that is not the method at all. Their capital, which determines what size is survivable. Their costs, which may be an order of magnitude below yours. Their tolerance for an ugly stretch. Their attention. Copy the method exactly and you have copied the smallest part.

There is also a selection problem sitting behind the whole industry. Out of a large population of traders taking risk, some will have excellent multi-year records through variance alone. Those are exactly the people with the credibility, confidence and material to start teaching. The market for mentorship selects on track record, which is a noisy signal, and it selects hardest at the moment the record looks best, which is when it is most likely to be at its most flattering.

What actually makes a method profitable

Strip the vocabulary away and there is one equation underneath everything.

Several things follow immediately, and most retail discussion ignores all of them.

Win rate on its own is meaningless. A method that is right nine times in ten and loses twenty times more when wrong is a losing method. A method right one time in five can be excellent. Any claim built around accuracy alone has told you the least informative of the four numbers.

Costs are not a detail. Commissions, exchange fees, the spread, and slippage all subtract from expectancy directly, and their effect is proportionally largest on the shortest-horizon methods, which are the ones most commonly sold. A method with a genuine but small edge is turned into a losing one by costs alone, and this happens far more often than anything exotic.

Variance decides what you experience. Two methods with identical expectancy but different variance produce entirely different lives. The higher-variance one will hand you drawdowns long enough to make you abandon it before its expectancy is realised. Position sizing is what converts an expectancy into a survivable path, and it is the part of the subject with the least glamour and the most consequence.

Everything above is non-stationary. The four numbers are estimates from a past period, drawn from a market whose participants have since changed. Nothing guarantees they are still the numbers.

What makes a method unprofitable

Most failures are not exotic. They are a short list, repeated.

• Overfitting. Adding parameters until the historical curve looks good. Every additional setting increases the chance the result describes the specific past sample rather than the market. A method with nine settings that performed beautifully has usually been sculpted, not discovered.
• Multiple testing. Try two hundred variations and keep the best one, and you have found the luckiest of two hundred, not the best of two hundred. Unless the search itself is accounted for, the winner’s historical performance is an overstatement by construction. This is the single most common way a sincere researcher fools themselves.
• Look-ahead bias. Using information in a test that would not have been available at that moment. Session closes, revised data, index membership, and anything computed across the whole sample are the usual culprits.
• Survivorship bias. Testing on the instruments that still exist. The ones that were delisted, expired worthless or went to zero are missing from the data, and they were the losses.
• Ignoring execution. Assuming fills at the touch, ignoring queue position, assuming size available that was not. Backtests fill perfectly. Markets do not.
• Too small a sample. Thirty occurrences tells you almost nothing about a process this noisy. Confidence should scale with count, and it usually scales with vividness instead.
• Regime dependence. A method developed entirely inside one volatility regime, one rate environment or one trending stretch has been tested against one draw from the distribution, not against the distribution.
• Crowding. The method worked, it was published, and enough participants acted on it that the opportunity closed. Nothing was wrong with the original research.

How to read an offer

Not a checklist for spotting fraud. A list of questions any honest seller should find easy, and which the structure of the industry makes awkward.

• What is the capacity of this method, and what happens to it as subscribers are added?
• What was the worst drawdown, how long did it last, and what did the equity curve look like through it?
• Are the results independently verified, from a broker statement rather than a screenshot, and do they include costs?
• How many variations were tested before this one was selected?
• What market conditions was it developed in, and what has it done in conditions unlike those?
• How does the seller earn money, in full, including rebates and affiliate arrangements?
• What would falsify the claim? If nothing would, it is not a claim about markets.

Where this leaves education

None of this means nothing can be taught. It means a distinction is worth holding on to.

Mechanics can be taught. What an option is. How margin works. What a footprint cell counts and how the classification behind it is inferred. Why two feeds disagree. This is knowledge, it is verifiable, it does not decay when more people know it, and there is no reason it should cost anything much.

Edges cannot really be sold. Not because sellers are all dishonest, but because of what an edge is. It is relative, depleting, capacity-limited and context-dependent, and the parts that make it work in one person’s hands are largely the parts that cannot be written down.

Which is the reason this site publishes the first and refuses to publish the second. Not modesty, and not caution. It is the only intellectually honest position available. Everything here is meant to help you understand what you are looking at. What to do about it is a research problem, and it is yours.

The Column makes the same argument in shorter form.