Swartberg Capital is a systematic research and trading firm in South Africa, owned and run by its two principals. The principals design the research, read the evidence, and make every decision that matters. Artificial intelligence assists them, and it gives them the capacity to apply that discipline to more ideas, with more consistency, and with less of the day lost to routine. What we publish here describes how the firm works.
The book draws on nine strategy sleeves, each grounded in published research and each drawing on a different economic link. The weight of each sleeve follows the regime the market is in. Below, an example market moves through four regimes and the book adjusts with it. It is an illustration of method, not market history.
Swartberg Capital exists to find, in the published record of how markets behave, what survives testing on our own data, and to trade only that. The work is quantitative: every claim is tested, every outcome is measured, and nothing is asserted that the data has not supported. The judgement that directs the work belongs to the principals, and artificial intelligence extends their capacity to exercise it. What cannot meet the standard is removed, not softened. This is the way the firm is managed.
Three disciplines, in that order. Each is applied by the principals to every idea the firm considers, and each answers to the evidence.
We begin with what has been published about how markets behave, and we keep every source. The principals decide what is worth reading and what enters the firm's thinking, and nothing enters without a citation that can be opened.
We test what we read against our own data, in the same way every time, under rules written down before the test begins, with the number of attempts kept on record. A result is measured, never asserted.
Limits are set before anything else, by a function that does not do the research, and the principals are responsible for it. The size of a position follows the evidence and nothing else.
What the firm knows, it knows because one of its principals judged the evidence sufficient, and what it does not know is said plainly. That judgement is not delegated.
The method is written down and applied the same way each time, to every idea, in the same order. A result is accepted on the evidence for it and on nothing else, and the record of what was tried travels with it.
Artificial intelligence assists the principals. It allows the same discipline to reach more ideas, more of the published record, and more of the day than two people could cover alone, and it returns the hours that routine would otherwise take. The research remains theirs, and so does the decision.
We publish what we learn in doing this work well. The notes carry no performance, no forecasts, and no recommendations.
What has to be true before the firm will say a result holds, and why the number of things we tried is part of the answer.
Much of what appears to be a result on a chart is returned to the market in trading costs. How we measure the cost of a trade before we count anything.
The firm's research rests on 206 published sources. How we verified each one, what we corrected, and why anything that could not be confirmed was removed.
Academic and industry research, including working papers, journals, and central bank publications. We take the finding and the mechanism behind it, not a trading rule, and the principals decide which findings merit the firm's time.
Entry, exit, and position size are written precisely enough to be coded, so that the test and the trade follow the same rule. Size is set by a rule decided in advance, not adjusted on the day.
Every simulated fill is taken at the next price that was actually available, with the spread quoted at that moment, commission, and any financing cost. No test is allowed a price that could not have been had.
Each rule must pass a sealed holdout period it never saw, walk-forward testing by year, cross-validation designed to limit leakage, a correction for the number of attempts, permutation tests for any machine-learned filter, and cost stress above the quoted spread. The principals read the evidence and decide whether the result holds, and a rule that fails any one of these is set aside.
Each line of research is tagged with the conditions in which its mechanism is present. A daily assessment, reviewed by the principals, decides what is eligible, and anything eligible but outside its own tested distribution is paused.
Each rule runs on a demo account first, because slippage is measured, not assumed. It goes live only once the fills have been measured, the result remains inside its validated distribution, and the principals have decided that it should.
Swartberg Capital is a quantitative firm, and its people are its edge. Artificial intelligence assists the principals in that work. It gives them capacity: the discipline set out on this page can be applied to more ideas, with more consistency, and with less of the day lost to routine. It does not set the question, weigh the evidence, or take the decision. Those remain with the principals, and every figure they see can be traced to the record it came from. The purpose of the technology is to give the principals more of their own time for the work that only they can do.
Research proposes, validation tests, risk sets the size, execution fills, and the principals decide. Each function is kept separate from the others, every recommendation carries its provenance, and the decision at every gate is a person's. This is the way the research is run.
The research behind the firm rests on 206 published papers, regulator texts, standards, and vendor documents. Every one was opened and checked against the claim it supports. All were found, a small number of dates and volume figures were corrected, and the rule stands that anything that cannot be confirmed is removed, not softened.
A result has to survive more than one test before the firm will say it holds: a sealed period the rule never saw, costs at the level actually quoted, and a correction for how many things were tried before this one. The number of attempts is part of the answer, and we keep that count on record.
Much of what appears to be a result on a chart is returned to the market in trading costs. We measure the cost of a trade before we count anything, at the spread quoted at that moment rather than an average, and we test what happens when fills are worse than quoted.
Some bars only print once price has already moved through a level. A test that fills an order at that level is describing a trade that could not have been made. We simulate fills at the next available price instead, and the correction changes the conclusion.
A modest result supports only a modest position, and sizing beyond it turns a sound rule into a wager, however good the statistics appear. Position size is decided by the evidence, by a function separate from the research, and it is capped. The principals answer for that function.
Features built from price contain less information than is commonly hoped. Our own tests reach that limit quickly, and the permutation tests say so. The next unit of understanding comes from new information, such as the calendar, related markets, and the order book, not from a larger model. Deciding where to look next is a judgement, and a person makes it.
Markets keep local hours, and the clock shifts twice a year. A test anchored in universal time samples part of the year and treats it as the whole. It is a small detail that changed a conclusion, which is why every rule we write that refers to the clock is anchored in the market's own time.
Swartberg Capital is a systematic research and trading firm in South Africa, owned and run by its two principals. It is a quantitative firm at heart: claims are tested against our own data, outcomes are measured, and nothing is asserted that the evidence has not supported. The principals are the firm's edge. They design the research, read the evidence, and make every decision that matters, and artificial intelligence assists them by giving them the capacity to apply that discipline across more ideas with the same care. Data, models, and results are kept under version control, with every figure reproducible from the underlying tick data. This is the way the firm is owned and run.
Swartberg is the black mountain range and pass of the Western Cape, a name in the same register as other South African investment firms. It was chosen from a screen of 85 candidates for domain availability, existing investment firms, company registers, and connotations. The .com and .co.za domains were available at the time of the screen. The name would be adopted formally only once a trademark search and a counsel clearance opinion are complete. The codename Quantum Fund was set aside because it is the name of George Soros's fund, still in use, and the word is a registered financial mark in several registers.
This site shows no performance. Should the firm ever present results, it would do so with the disclosures a regulated manager would use: hypothetical results labelled as such, costs stated, and no simulated figure described as a return earned. A live record would be verified by an independent third party before it was shown to anyone, and the principals would stand behind every figure in it.
[Date]
Dear colleague,
Thank you for your time this week. The note we discussed is attached. It describes how the firm works, the standard its principals hold themselves to, and the place of its people and its technology in that work.
Swartberg Capital does not offer products or services to the public and does not accept capital from any other person. Correspondence about the firm's research, its standards, and the way it works is welcome, and it is read by the principals.
Jared van Heerden
Director
jared@swartbergcapital.com
Jason Biddulph
Director
jason@swartbergcapital.com
South Africa. No office address is published.