1. Market regimes: when a strategy stops working

Macro 6: From macro to strategy

The previous five levels have been assembling loose parts: the price of money, who sets it, how it travels along the curve, what happens when inflation gets out of hand and how credit breaks. This level puts them together, and it starts with the idea that saves an investor more money than any other: the market does not have one single physics, it has several, and it switches between them without warning.

A regime is not a fashion, it is a stable set of rules

We call a regime a long stretch during which the relationships between variables hold reasonably steady: the correlation between equities and bonds, which sort of company gets rewarded, how the market reacts to a surprise in a data release, how long a drawdown lasts before it recovers. Inside a regime the market appears to obey laws. Strategies fitted to those laws work, and whoever designed them tends to take the credit.

The trouble is what happens on the other side. When the regime changes, the strategy does not break: it carries on executing exactly as well as before, buying and selling precisely what you told it to buy and sell, and losing money month after month. There is no bug to fix in the code or in the rule. The mistake is that the rule encoded a property of the environment, not a property of the market.

Three regimes worth keeping in mind

• 1982-2021: four decades of falling rates. The yield on the ten-year US Treasury had been above 15% in the early eighties and was hovering around 0.5% by the summer of 2020. For almost forty years, any portfolio with long duration enjoyed a permanent tailwind that owed nothing to anybody's skill.
• 2009-2021: zero rates and expanding central bank balance sheets. Level 2 explained the mechanism. What matters here is who it rewarded: growth without profits, long duration, cheap leverage, buying every dip and selling volatility. And who it punished: cash, caution and any form of hedging, which for thirteen years was a cost with nothing to show for it.
• 2021-2023: the return of inflation. US CPI went above 9% year on year in June 2022 and the Federal Reserve raised rates by some 425 basis points in nine months. Everything the previous regime had rewarded became a liability, and it happened within months.

Why a strategy validated inside one regime is not validated

There is a statistical illusion here that fools even well-trained people. Thirteen years of daily data look like some three thousand observations, which sounds like an enormous sample. It is not. It is three thousand heavily correlated details of one single observation of how one single macro environment behaved. To know whether a strategy survives a change of regime you need to have watched it live through several of them, and those are counted on the fingers of one hand, not in thousands of candles.

Marcos López de Prado insists on a complementary point that is even less comfortable: the more variants you try on the same history, the better the best result you will find by pure chance. If you test four hundred parameter combinations and keep the winner, you have not discovered a rule, you have selected the largest stroke of luck available in that sample. And if that sample covers a single regime on top of it, the result does not even describe the market: it describes an era.

Before accepting any validation as sound, it is worth answering three questions in writing:

• Which macro regime does my history cover, and how many genuinely distinct regimes does it contain?
• What would have to happen in the environment for this strategy to stop working?
• How many variants did I try before settling on this one?

The market does not ring a bell

Nobody puts out a statement announcing that the regime has changed. The changes get dated years afterwards, once they have a name, and in real time they turn up disguised as a bad patch. That is the cruel part: during the first months of a break, the losses usually fall inside the strategy's historical range of drawdowns, so every alarm you set will tell you to sit tight. And it will tell you so precisely because that range was calculated with data from the old regime.

Telling a bad run apart from a structural break in the moment is, statistically, next to impossible: to be certain you need so much data that by the time you have it you will have paid for the lesson in full. Howard Marks puts it in a way worth adopting: you cannot know where you are going, but you can know where you are. The point is not to anticipate the change, but to recognize which environment you are trading in and what that environment is rewarding.

A real case: Japan, thirty-four years

The Nikkei 225 closed on December 29, 1989 at 38,915 points, after a decade in which Japanese equities were the best performing asset on the planet and in which people explained quite matter-of-factly why the Japanese model justified multiples seen nowhere else. What followed was two long decades of deflation and deleveraging. The index fell below 8,000 points in 2003 and again after the 2008 crisis, and it did not regain that high until February 2024: thirty-four years.

A Japanese investor who in 1989 had validated any strategy on the previous twenty years would have concluded, on impeccable data, that buying the dip always works and that the stock market doubles every few years. Both statements were true inside their regime and false outside it. The belief that every fall recovers within a couple of years is not a law of the market: it is a description of the American experience of recent decades, which is exactly the sort of thing a regime produces.