Sector rotation: would seeing it coming make you rich?
- Robin Powell
- 9 minutes ago
- 9 min read
The past six weeks have brought a sharp sector rotation in the US stock market, the kind of regime change that tempts investors to believe a clever forecaster could have called it and cashed in. The evidence points the other way. Even handed tomorrow's headlines in advance, most people — and, it turns out, most machines — struggle to turn foresight into profit.
Since the middle of May, a sector rotation has been quietly under way in the US market. The megacap technology stocks that had carried the index for years stalled, and the money that left them moved beneath the surface: first into semiconductors, then into the more defensive corners of the market. The index itself barely moved, holding roughly flat rather than falling. Robert Armstrong, who writes the Financial Times's Unhedged column, has tracked the shift in real time, floating one explanation after another — worries about AI spending, ordinary profit-taking, a bet on interest rates — and settling on none. By his own account, the semiconductor rally had already begun to fade.
The natural reaction is to assume that a sharp enough observer should have seen the turn coming and positioned for it ahead of the crowd. Sector rotations look obvious once they have happened, and the explanations arrive promptly to tell us why they were inevitable. Yet the people who forecast markets for a living have a poor record of calling turns before they happen.
But the problem runs deeper than the difficulty of forecasting. Suppose you had known. Suppose the whole sequence had been handed to you in advance. The evidence below suggests that even perfect foreknowledge of the news would not reliably have made you money, and the reason has little to do with the quality of the forecast.
Tomorrow's news, in advance
A recent piece in the Economist revisited an experiment built to test an old idea: that an investor given tomorrow's news would still find a way to go broke. The conjecture belongs to Nassim Nicholas Taleb, who argued in 2016 that a day's advance sight of the headlines would ruin most investors inside a year. Victor Haghani and James White of Elm Wealth decided to put it to the test.
The setup was deliberately generous. In November 2023, Haghani, White and their colleague Jerry Bell gave 118 finance-trained adults — more than 90 per cent of them studying for graduate finance degrees or MBAs at four US east-coast universities — $50 each and a crystal ball. The ball was the front page of the next day's Wall Street Journal, handed over a full day early with the market-price data blacked out. Players could trade the S&P 500 and 30-year US Treasury bonds, with up to 50 times leverage, across 15 trading days drawn one per year from 2008 to 2022. To make the advantage worth having, the days were taken from the more volatile half of the record: employment reports, Federal Reserve announcements and the like. Positions opened at the previous day's close and settled at the following day's close.
With tomorrow's front page in hand, the players ought to have cleaned up. They did not. About half of them lost money. One in six went bust altogether. The average participant walked away with $51.62 on a $50 stake, a gain of 3.2 per cent that Elm describes as statistically indistinguishable from breaking even. Across roughly 2,000 trades, the players called the direction of stocks and bonds correctly just 51.5 per cent of the time, barely better than a coin.
The experiment has since been turned into a free online game that some 60,000 people have played, and on Elm's account they have fared substantially worse than the original 118, though that is a separate, self-selected crowd and the two groups are not directly comparable.
Being right was not the problem
Why could a group of finance students, holding the next day's news, barely break even? The answer comes in two parts, and the less obvious part matters more.
The first part is simply that their forecasts were poor: a 51.5 per cent hit rate is not much of an edge. The second part reframes everything. The players had no discipline about how much to stake. Elm found that the size of a player's bet bore essentially no relation to how likely it was to pay off: the correlation between the two was about zero for stock trades and slightly negative for bonds. They did not bet more when they were more likely to be right. And they leaned hard on the leverage on offer, gearing up beyond 20 times their capital on roughly a third of the days they traded, beyond 60 times on about one day in 25, and reaching the full 100 times on a handful of occasions. At that exposure, a single day moving against them ended the game.
To separate sizing from foresight, Elm ran the same game past five seasoned macro traders, a select group that included the head of trading at a top-five US bank and a former senior trader at Jane Street. They did far better. Their average return was 130 per cent, with a median of 60 per cent; the best turned a stake into more than five times its starting value, and none went bust. Yet their forecasts were only marginally sharper than the students': they called direction correctly about 63 per cent of the time, against 51.5 per cent. A gap that small in prediction produced a far larger gap in outcome, and the reason was discipline. The experts staked selectively, sitting out roughly a third of the opportunities and betting heavily only when they were confident. 'Deciding how much, it seems, is tougher' was how the Economist summarised the lesson.
There is a mathematical reason why a good forecaster can still lose money by staking too much, and a later section sets it out.
What happened when the machines played
<!-- VISUAL: In-article image — a hand hovering over four stacks of coins of markedly different heights; the harder decision is how much to commit, not which way. Placed at the top of this section. Generated in the editorial pass. -->
Earlier this year Elm asked the obvious follow-up: what would the machines do? Bell, Haghani and White gave the same crystal ball to four leading AI models, running each through ten rounds of the game with a starting pot of $1 million and an instruction to behave like a typical middle-aged investor with their whole wealth on the line.
On the part of the task that looks hardest — reading the news and calling the market — the better models did about as well as the expert humans, with top hit ratios around 60 per cent.
On the part that looks easier, they failed exactly as the students had. They bet far too aggressively for the goal they had been set, running daily swings in the value of their portfolios of between 20 and 40 per cent, several times wider than their edge could justify. Two of the four made money on average; but across only ten rounds each, the winners were as much lucky as skilful. For perspective, Elm notes that the US stock market has moved by more than 5 per cent on 23 days, and by more than 9 per cent on seven days, since 2000. At the position sizes the models were running, that kind of move could wipe out an account.
Yet the models knew better. Asked to size a bet on a simple coin-flip game, every one gave the textbook-correct answer. They could state the principle in the abstract and then fail to apply it the moment the same problem was dressed in market clothing. When the researchers had the models 'read' Haghani and White's book on financial decision-making before playing, their staking improved sharply, while their forecasting did not move. The thing that got better was the thing that had been going wrong: not the prediction, but the size of the bet.
With sector rotation, the size of the bet decides the outcome
Step back from the experiments and the principle is straightforward, if counterintuitive. For any favourable bet there is an optimal amount to stake, and betting beyond it does not merely add risk. Past a point it starts to eat the returns themselves, so that being right most of the time is no longer enough to keep you moving forward.
The arithmetic is brisk. Say you have a bet you will win 60 per cent of the time at even money, a real and sizeable edge. The stake that maximises your wealth over the long run is about 20 per cent of your capital. Stake far more, half your capital each time, and even a good run of six wins in ten leaves you down by roughly 29 per cent, because the four losses cut deeper than the six gains restore. Stake around twice the optimal amount and your long-run growth rate falls all the way to zero. The mechanism is volatility drag: raising your bet size lifts expected return in a straight line but lifts risk faster, and beyond the optimum the drag overwhelms the edge. This is why the crystal ball disappointed and why the machines stumbled. Knowing the direction is only half of the decision, and the more dangerous half is how much to commit.
The point lands hardest against sector rotation itself, which is at bottom a form of market timing, and the evidence on timing the market is consistently discouraging. Alexander Molchanov and Jeffrey Stangl of Massey University in New Zealand examined how industries actually performed across ten business cycles dated by the National Bureau of Economic Research, from 1948 to 2018, and asked what an investor blessed with perfect foresight of every turn could have made by rotating between sectors accordingly. The answer, in their study, was a risk-adjusted edge of about 0.11 per cent a month before costs. Once realistic transaction costs or small timing errors were allowed for, even that slender margin dissolved into something statistically indistinguishable from zero. Sector returns, they concluded, show neither systematic nor persistent differences across the stages of the cycle. A flawless forecast of the sector rotation, in other words, was worth almost nothing once the real world was let back in.
The same mistake, closer to home
None of this is an abstract or an American problem. UK retail investors are making the same error right now: staking too much, and holding geared products designed to be owned for a single day far longer than that.
In January 2026 the Financial Conduct Authority published a review of how complex, leveraged exchange-traded products are sold, and it put numbers to the behaviour. The number of retail consumers trading these products had risen 23 per cent in a year. Triple-leveraged products are among the more popular on the London Stock Exchange. And of 531,007 trades in complex ETPs that the regulator examined, 82 per cent were held beyond the single day the products are built for. These instruments reset daily; held longer, they decay through tracking error and volatility drag, the same compounding penalty that sinks an oversized bet.
Which brings the argument back to where it started. The experiments point to a conclusion that is easy to state and hard to live by: the binding constraint on most investors is not information but discipline — the willingness to size a position sensibly and to sit out when the edge is not really there. Elm's own reflection on its AI results runs along the same lines. The authors suggest that artificial intelligence may prove most useful not as an autonomous trader at all, but as a counterweight to the human urge to bet too big on a view that feels certain and is not, even though other research finds AI can amplify our biases as readily as restrain them. That is a quieter role than the one usually imagined for it, and a more plausible one. It is also, more or less, the job a good adviser already does.
Armstrong, surveying the sector rotation he had spent six weeks trying to explain, allowed that 'this market feels adrift to me'. He may be right. But the lesson of the crystal ball is that even a clear view of where the market is heading would not tell you the one thing that decides whether you come out ahead: how much to wager on being right.
Resources
Bell, J., Haghani, V., & White, J. (2026). Do AIs make good traders, and do they make good traders better? Elm Wealth.
Financial Conduct Authority. (2026). Complex exchange traded products: good practice and areas for improvement. Financial Conduct Authority.
Haghani, V., White, J., & Bell, J. (2024). When a crystal ball isn't enough to make you rich. Elm Wealth.
Kelly, J. L. (1956). A new interpretation of information rate. Bell System Technical Journal, 35(4), 917–926.
Molchanov, A., & Stangl, J. (n.d.). The myth of sector rotation. Auckland Centre for Financial Research, Massey University.
An adviser as a check on the urge to bet too big
The closing thought of this piece is that the binding constraint is discipline, not information. It describes the part of investing a good adviser is there to supply. For readers who suspect their own instinct is to bet too big when a view feels certain, TEBI's Find an Adviser directory lists advisers who have publicly committed to evidence-based investing, and who regard a check on that instinct as part of the job.
For readers who would rather work through the thinking themselves first, How to Fund the Life You Want by Robin Powell and Jonathan Hollow sets out how to make sound decisions with money you have worked hard to build. Bloomsbury published the second edition; it is available on Amazon.
