Growth, Inflation and the Four Regimes

The four quadrants and what each one pays

Take growth above or below trend on one axis and inflation above or below target on the other. Above trend with inflation contained is the goldilocks quadrant: equities and credit work, duration is carried rather than owned, and the risk is that the regime is priced as permanent. Above trend with inflation above target is reflation: nominal assets do best, including commodities, real assets, cyclical equities and banks, while long bonds are the funding for the trade rather than a holding. Below trend with inflation contained is disinflation: long duration is the winner, both in bonds and in the long-duration equities that hurt most on the way down, and the risk is that the slowdown becomes a recession. Below trend with inflation above target is stagflation, and it is the quadrant with no easy winner: equity multiples compress because the discount rate is high, bonds lose in real terms, and the assets that hold are real assets, cash-generative businesses with pricing power, and short-duration credit. The transitions are what move prices, and they are not symmetric. A shift from goldilocks into reflation raises nominal growth, which helps earnings and hurts multiples; a shift from reflation into stagflation raises the required real rate, which hurts both bond and equity valuations at once. The most damaging transition for a conventional portfolio is the move from a low-inflation regime into a high-inflation one, because it is the only one in which the two halves of the portfolio are driven by the same variable in the same direction — the discount rate — which is exactly what made 2022 a year in which both legs lost. Reading the quadrant requires an estimate of trend, which is the hard part. Growth above trend is not the same as growth above the previous quarter; it is growth above what the economy can sustain without generating inflation, and that is estimated rather than observed. The same is true of the inflation target, which is 2% for most developed-market central banks but which is a policy choice rather than a fact. The practical approach is to use a small number of proxies — real growth against a moving average of potential, core inflation against target, and the direction of both — and accept that classification is a judgement with a margin of error rather than an observation. Four quadrants, four winners — Growth above, inflation contained: equities and credit; duration carried · Growth above, inflation above: nominal and real assets, cyclicals and banks; long bonds lose ← · Growth below, inflation contained: long duration: bonds, then long-duration equities · Growth below, inflation above: pricing power, short duration, real assets; no easy winner A regime is not a forecast. Being able to name the quadrant does not tell you when it changes, and the empirical record of regime-timing strategies is mixed at best. What classification buys is an audit of what you own and why: if the assets are the ones the current quadrant pays, that is a position rather than a default.

Why the correlation is the regime

The reason a portfolio built for one quadrant fails in another is not that the assets changed, but that the relationship between them did. Equities and government bonds were negatively correlated for most of the two decades before 2021, which made a balanced portfolio smoother than either leg — and that correlation was not a law of nature, it was the property of a regime in which demand shocks dominated, so a weak economy lowered both the discount rate and expectations, and the two assets responded in opposite directions. When supply shocks dominate, the sign flips: a higher inflation print raises nominal rates, which lowers bond prices and compresses equity multiples at the same time, and the diversification disappears exactly when it is wanted. That is the argument for holding assets that are chosen for different quadrants rather than for a mix chosen by historical volatility. A portfolio that holds equities, long bonds, real assets and cash-like instruments is not diversified because their average correlations were low; it is diversified because the four are supported by different quadrants, so one of them is doing well in each state. Which of these you need depends on the liabilities, on the horizon and on tolerance for tracking error against a benchmark that is itself a single-quadrant portfolio. There is a practical caution about the length of history used to estimate any of this. Correlation estimates are dominated by the largest events in the sample, so a window that contains only demand shocks will report a negative stock-bond correlation with a high degree of confidence that is really a statement about the sample. Longer windows that include the 1970s report the opposite sign, and the honest conclusion is that the sign is a regime variable with a plausible range spanning zero — which is why the regime classification is more useful than the correlation matrix. • The stock-bond correlation is a regime variable, not a constant. • Demand shocks make the two move oppositely; supply shocks make them move together. • Hold assets chosen for different quadrants rather than for low historical volatility. • Correlation estimates are dominated by the largest events in the window used. The 2022 experience is the reference case: the highest inflation in four decades arrived with an aggressive tightening cycle, and a 60/40 portfolio lost about 16% — the loss concentrated in the one correlation that the whole construction assumed away.

You will always recognise the regime late

The four-quadrant framework is genuinely useful and it has a timing problem that no amount of care removes. Regimes are identified from data that is **published with a lag and then revised**, and a turning point is by definition the period when the data is least informative — the growth numbers are mixed, the inflation prints are noisy, and the series that will later define the period have not finished being revised. The classification of a quarter as the start of a new regime is usually made several quarters after the fact, when the growth and inflation series have settled. A framework that tells you where you have been is still useful, but only if it is used for something other than timing entries. The second problem is the cost of switching. A rule that reclassifies the regime frequently will trade on noise, paying the spread and the tax consequence each time, and its apparent responsiveness is a cost rather than a feature. A rule that reclassifies rarely will be slow at genuine turns, which is tolerable if the response to a regime change is **sizing** rather than a wholesale switch — reducing exposure to the assets the previous quadrant rewarded and adding to the ones the new quadrant favours, in stages. The framework earns its keep as a way of choosing what to own more of and less of over quarters, not as a signal to change the portfolio on a Tuesday. The third problem is the one worth writing down before it happens: the data that defines a regime is also the data the market has already seen. By the time a quadrant is unambiguous, the assets that the quadrant supports have usually moved. That does not make the framework useless — positioning for the persistence of a confirmed regime is a legitimate strategy, and regime persistence is real — but it does mean that the edge is not in the identification. It is in the positioning and in the sizing, which is exactly where the risk lessons put it: you cannot be early with information you do not have, and you can only be wrong in a survivable way. • Data arrives late and gets revised, so turning points are identified in retrospect. • Frequent switching trades on noise; the response to a regime change should be sizing, not an all-in switch. • By the time a quadrant is clear, the market has seen the same data — the edge is not in identification. • Regime persistence is real, which is what makes positioned-for-it exposure legitimate.

The dashboard: which series, in which order, and the second derivative

Classification is only as good as the series behind it, and the choice is narrower than it looks. For the growth axis, the most informative instruments are a **survey of purchasing managers** — the manufacturing and services diffusion indices, in which a reading above fifty means more respondents report expansion than contraction — a **jobless claims** series, which is weekly, high-frequency and one of the least revised statistics published, and a **coincident activity index** to say where the economy actually is rather than where a survey thinks it is heading. The growth axis is therefore read forward-looking from surveys and claims, and confirmed backward-looking from output and employment. For the inflation axis the order is different: a **core** measure that strips food and energy to avoid reading a commodity price as a regime, the **inflation-linked bond market’s breakeven** as the traded expectation, and a **wage or unit-labour-cost series** as the slow variable that decides whether a shock becomes persistent. Two of those three are available monthly, one is available continuously, and none of them is precise — which is why the dashboard should be read as a scorecard with a score, not as a single reading. The second rule is that the dashboard is read for **direction**, not level, and the directional reading is the second derivative rather than the first. Growth above trend but decelerating is a different quadrant from growth above trend and accelerating, even though a level-based reading puts both in the same box — and the transition is what reprices portfolios, so the change in the change is where the analysis lives. Practically that means comparing the latest three months against the prior three rather than against a long-run average, watching whether a series that has been improving is improving more slowly, and treating a survey that is still above fifty but falling for four consecutive months as a growth-negative signal. The same logic applies to inflation: a core rate of three percent that has fallen for five months is a different regime from three percent that has risen for five, and the level alone cannot distinguish them. The third rule is about the data’s own quality, which the previous read already flagged and which changes how the dashboard should be used. Release lags are uneven: claims arrive weekly, inflation monthly, output quarterly and revised for years. Surveys are noisy month to month and should be read as a three-month average. And the series that will eventually define a regime are the ones most heavily revised, so a reading taken in real time is systematically less informative at turning points than in the middle of a regime. The remedy is not to wait, which costs the opportunity; it is to hold two readings — where the dashboard says the economy is now, and where it says the economy will be in two quarters — and to size positions by the confidence in the second rather than the certainty of the first. That framing is what makes this lesson’s output an audit rather than a prediction: the dashboard tells you which quadrant your holdings are built for, and the second derivative tells you whether that quadrant is still arriving. The dashboard, in order — Growth, forward: Purchasing-manager surveys and weekly claims — read as three-month averages · Growth, confirmed: Coincident activity and employment — backward-looking, less revised ← · Inflation, core: A core measure plus a wage or unit-labour-cost series for persistence ← · Expectations: The traded breakeven, read against its own range rather than a survey alone The output of the dashboard is one of four labels plus a confidence, and the confidence is the part most people drop. A reading taken when the four series agree is worth acting on at normal size; a reading taken when surveys and coincident data disagree is worth acting on at half size, because the disagreement is itself the statement that the regime is in transition.

Sources

• Growth and inflation regimes and asset returnsStandard regime-based asset-allocation research; Bridgewater’s “all weather” framing; Ilmanen, “Expected Returns”
• Correlations between stocks and bonds as a regime variableStandard evidence on the stock-bond correlation in inflation and disinflation regimes
• Real rates, inflation shocks and durationMR13 on duration; standard literature on inflation and asset valuations