Strategies rarely break — the market stops supplying what they require, and the unit of account is the one position nobody hedges
Investment strategies die with striking regularity, and the obituaries almost always name the wrong cause. The logic rarely breaks; the market stops supplying what the strategy requires. Every strategy is a machine built for specific market weather — call the set of conditions it was built for its operating envelope — and the useful question about any of them is not which regime comes next, which nobody knows, but what this one needs and whether that is currently on offer. Specification, not prediction. It can be answered from data in hand.
The essay states the question properly, and then the two claims that follow from taking it seriously. First: market states should be defined by six observable properties of prices rather than macro labels — a deliberately parsimonious map — and once a strategy's harvested property is named, its envelope, and the way it dies, can be specified to first order. Second: the largest position in every portfolio is one nobody hedges — the unit it is measured in.
Three deaths, from one recent decade. Intraday equity mean reversion — buy the dip, sell the pop, hundreds of times a day — was a reliable business through the 2000s and stopped paying after about 2010. Nominal government bonds hedged equity portfolios for twenty years and then, in 2022, fell together with the stocks they were held to protect. Short-volatility strategies compounded quietly through the 2010s and returned years of premium in single sessions when the calm ended. Each had a decade or more of working first. In none of the three did the logic break.
The label to strip is "strategy death." A strategy is a machine built for specific market weather. A trend follower is a machine for harvesting moves that continue; a carry trade is a machine for harvesting calm; neither is good or bad, and neither works everywhere. Call the set of conditions a strategy was built for its operating envelope: the states in which its return engine runs, outside of which it bleeds or breaks. What dies is almost never the machine. What moves is the weather.
Farmers understood this before finance did. A strategy is a crop with growing conditions, and the questions worth asking are agricultural: what does it need, is the ground currently supplying it, and how does it die — slowly, in drought, or all at once, in frost?
Nobody asks whether wheat is good. They ask whether this is wheat country.
This essay builds that question properly: what does each strategy require, and is the market supplying it. Answering either half requires no forecast of anything.
Before diagnosing any death, fix what "died" means — died in which unit. Every return ever printed is a ratio: an asset over a unit of account. The numerator gets all the attention; the denominator is a position. Choosing a numéraire looks like a measurement convention. It is the largest allocation in the portfolio — a 100% position, unhedgeable, held permanently, in the one asset that cannot be diversified: the unit wealth is counted in. It is the choice of what counts as the harvest. The exposure is conceptual, not a cash balance — a dollar-based investor has not bought dollars by reporting in them. What the unit fixes is the definition of gain and loss itself, which is why it belongs among the allocation decisions rather than the conventions.
The choice defines what counts as riskless for the holder, and with it every other label. In dollars, Treasury bills are the zero-volatility asset and gold a volatile speculation. Count in gold ounces instead, and physical gold becomes the riskless leg while bills become a structural short on monetary credibility — negative expected return in the chosen unit, with occasional payoffs in disinflationary busts. Nothing in the portfolio changed; the labels hedge and carry swapped places. Hold that axiom. The essay returns to it at the end, once the machinery exists to state its full consequence.
"Which strategy for which market" is usually asked as a forecasting problem: identify the coming regime, position accordingly. Asked that way, it inherits every failure mode of prediction. Asked the other way — for which states was this strategy built, and are those states present — it becomes specification: an engineering question with an engineering answer.
The existing literature offers two traditions, and neither asks it. The normative tradition — Dalio's All Weather at Bridgewater, built on the growth-and-inflation quadrant, is the reference case (Bridgewater, 2012) — refuses to forecast and balances risk across macro states, so that something works whatever arrives. Sound construction. But its states are macroeconomic — growth surprises, inflation surprises — while strategies respond to statistical properties of prices that macro labels only loosely track. Trend following does not care whether a sustained move is driven by disinflation or by war; it cares that the move sustains.
The switching tradition — regime-detection models descending from Hamilton's founding paper (1989) — tries to identify the current regime and rotate. Its difficulties are structural. Regimes tend to be confirmed late, precisely when switching is most expensive; modern history offers few genuine transitions to fit on, which invites overfitting; and whipsaw plus transaction costs erode the paper alpha. But the narrower point matters more than the critique: to decide whether an existing strategy is inside its envelope, no classification of the world into a handful of named macro regimes is needed at all.
What neither tradition supplies is a statement, per strategy, of the conditions under which its engine runs. That is the missing layer, and it is the same move the structuring frameworks published here make for instruments, turned on return streams: the claim ladder asks what a holder actually owns when a structure fails; the envelope asks when what the holder owns actually pays.
Define market states by what strategies actually touch: measurable properties of the price process, not macro narratives. Six give a working map — deliberately parsimonious rather than exhaustive.
Each is observable from data in hand — return series, options prices, turbulence measures (Kritzman & Li, 2010), margin schedules — with no forecast required. A "regime" in this framework is simply a persistent configuration of the six. The 2022 stock–bond correlation flip shows the discipline: correlation was never the regime — it was an output of the inflation state. The statistical layer sits downstream of the macro drivers, but it is the layer strategies touch.
Once a strategy's harvested property is named, its envelope can be specified to first order. The rows are stylised: each names the primary return engine, not every implementation or every driver of realised returns.
Two readings of the map. First, the popular labels are wrong: gold is not an "inflation hedge" — its envelope is the real-rate and credibility state, which only sometimes coincides with CPI. Second, 60/40 is not "dead"; its bond leg simply left its envelope when inflation volatility returned, exactly as the map predicts.
If the map must be compressed to a single axis, it is this: is the strategy selling insurance, or buying it? Fung and Hsieh proved the canonical case (2001): trend-following returns replicate a lookback straddle. Trend is, in effect, long an option on large moves — paying rent, in the form of whipsaw, through quiet times, for a payoff in the tails.
The generalisation is Cole's (2020). Concave strategies — carry, short volatility, credit, arbitrage — sell insurance: they earn steadily in the middle of the state space and lose discontinuously at its edges, short an option they may not know they sold. Convex strategies — trend, long volatility, tail hedges — buy it: they bleed in the middle and get paid at the edges. The test requires no macro model — regress a strategy's returns on market moves and read the curvature; only a return history is needed. The classification is of components, not clans: most strategies contain an insurance-like element somewhere in the state space, and the empirical task is to locate its sign — real trend books carry financing, turnover and whipsaw effects that blur the clean straddle.
A portfolio's true regime exposure is the sum of its constituents' convexity profiles — and most institutional portfolios are unknowingly concave everywhere.
Four terms make this precise. The operating envelope: the observable conditions under which expected payoff is positive. The failure boundary: the conditions under which it deteriorates materially. Failure geometry: the path and speed by which losses accumulate as the boundary is crossed. And the recovery mechanism: whether the strategy repairs itself once the state passes, or capital is impaired before it can. Convexity gives the sign at the boundary; geometry is how the position travels there — and two strategies with the same envelope are not interchangeable if their rules behave differently as the state exits. Mean reversion dies catastrophically because its rule adds as prices diverge — a structural break turns the book into a levered bet against the new state. LTCM is the type specimen, and the ruin lay in the position path, not in the concavity as such. Trend dies gracefully because its rule cuts: whipsaw is an annuity of paper cuts with no mechanism for ruin. Short volatility gives back in days what it earned in years; long volatility dies slowly and in plain sight, which is psychologically hard and structurally safe. The recovery mechanism completes the picture: trend re-engages by rule once the whipsaw passes; the short-vol book and the levered mean-reverter are impaired before any recovery arrives.
An allocation therefore needs the full set per strategy: where it earns, how it fails, and whether it comes back. Sizing follows the failure terms, not the envelope. Catastrophic paths are sized to survive their failure state; graceful paths can be sized to their envelope.
The map is stable — what each strategy requires does not change. The territory drifts. In the metaphor's terms: the crop's requirements are climate; what the field gets is weather — and sometimes the climate zone itself moves. Intraday mean reversion decayed because high-frequency competition compressed the reversal it harvested; the required state stopped occurring at exploitable size. Duration left its envelope when inflation volatility returned. Short volatility's envelope had been artificially widened by central-bank reaction functions and was repriced when they changed. In each case the farmer who blames the seed is misreading the map: the correct monitoring question is diagnostic, not predictive — are the properties this strategy requires currently observable?
The assumption doing the work is persistence, and it should be stated rather than smuggled. Diagnosis is actionable only because the six properties move slowly relative to the cost of reallocating: they shift over months, while the switching tradition's losses concentrate at exactly the transitions this framework refuses to trade. The same estimate then serves two uses with different tolerances. At design time, misreading the current state costs little, because the allocation is built to cover all states. At run time, the only permitted response is de-risking (rule 4 below), where the cost of a false alarm is bounded — forgone return — while the cost the switching models carry is not: wrong-way exposure at the transition. Diagnosis earns its distinction from prediction not epistemically but through the asymmetry of what is done with the estimate.
One more force moves the territory: occupancy. Envelopes are shared, and crowding narrows them — every farmer planting the same crop moves the price at harvest. At sufficient scale a strategy manufactures its own failure state outright, a mechanism the companion register documents case by case. The practical rule: a compressed premium is not just lower expected return; it is evidence the envelope is crowded and its exit narrow.
The honest summary: what this yields is not a solved selection problem but a systematic way of stating a strategy's requirements and monitoring whether they remain present — an ex-ante design discipline, not an ex-post switching rule. Map the envelopes, cover the state space, size by how each strategy dies, and let diagnosis — never prediction — trigger a defensible dynamic response: holding less.
Now the axiom from §II can be stated in full. Every envelope in the §V map is the envelope of a ratio — asset over numéraire — not of an asset. The map is unit-relative, and so is every verdict built on it: a portfolio is only "all weather", in Dalio's sense, with respect to a stated unit of account. Which makes the choice of unit the one envelope decision that cannot be diversified, hedged, or held tactically. It can only be made — and most holders never notice they made it. Two test cases follow: applications of the framework, not settled verdicts.
The gold numéraire is a regime assertion. Adopting the ounce says: the state I refuse exposure to is fiat debasement. That immunises, by construction, against the states gold covers — and inherits gold's own failure state at full size. The 1980–2000 configuration — positive and rising real rates, restored monetary credibility, disinflation — was a two-decade bear market for the unit itself: holding the numéraire earned nothing while equities compounded enormously in ounce terms. For the ounce investor, Goldilocks is the crash. The consequence for diversification is severe: over the post-1971 record, few assets have beaten gold across states — broad equities over long horizons, little else with consistency — and most conventional diversifiers show negative returns in ounces across that record. The ounce portfolio that follows is concentrated — the numéraire plus the short list of assets with positive expected return in that unit, plus trend, which as a long/short engine survives the change of unit. Concentration is not a flaw of the ounce portfolio; it is what the choice of unit logically entails.
Bitcoin fails the framework's own criterion, and the failure is informative. State the criterion first: a workable unit needs an envelope one can map, and semantics that do not presuppose its own victory — because a numéraire the holder expects to appreciate against all assets is not a unit of account but a portfolio wearing one's clothes. Bitcoin fails on both counts. Its own envelope is unmapped — fifteen years spanning few distinct macro configurations, with volatility so high that, measured in it, every diversified portfolio is dominated by the unit's noise. And adopting it asserts the one state in which it wins, sustained monetisation, as the baseline — under which everything else has negative expected return and holding only the numéraire is the consistent portfolio. The criterion's honest corollary cuts against gold too: purchasing-power stability is verifiable only at horizons no holder has, and gold's own record — 1971–1980, then two decades of decline — is not stable semantics inside a career. What its centuries supply is not verified stability but a mapped envelope. Every numéraire choice is therefore a regime assertion, gold included; the distinction between the ounce and Bitcoin is evidential, not categorical. A unit whose entire case is appreciation is a position. A unit with a long, mixed, state-mapped record is an assertion one can at least underwrite.
One portfolio, one unconditional unit. A book cannot make two different units its unconditional definition of wealth; it can only rank them. The tempting compromise — grow wealth in ounces, cap drawdowns in fiat — leaves the ranking unstated precisely where it binds, because the two units anti-correlate in the states that force the choice: a gold bull market is the ounce mandate's best state and the fiat-drawdown mandate's stress test at once. The honest resolutions are to rank the mandates explicitly, to measure the subordinate one over rolling multi-year windows rather than point-in-time, or to run two books with separate units. What cannot be had is two unconditional definitions of wealth in one book.
Two farmers work identical land. One counts the harvest in bushels, the other in what the bushels sell for. Same fields, same weather, different books — and in a bad year for the currency, only one of them is rich. That is the closing statement of the framework: envelopes can be covered in assets, never in units. The model says how to build an all-weather portfolio relative to a stated numéraire; the numéraire itself is the one weather you have chosen to stand in.
Written in a personal capacity. Analytical views only — not legal or investment advice.
Julian Gretzinger — Investor and writer on monetary history, real wealth mechanics, and financial markets. substack.com/@juliangretzinger