120 Years, 16 Countries, 4 Regimes: What the History of Global Markets Says About Inflation

Ray Dalio's framework sorts the world into four states by whether growth and inflation are rising or falling. Each asset, the argument goes, earns its return in a different quadrant. It is a tidy story, and it survives 120 years of data. Mostly.

That "mostly" is the point of this article. Building on Beyond Passive Investing's analysis of the Jordà-Schularick-Taylor Macrohistory Database (Release 6, covering 16 developed markets from 1901–2020), we replicate the framework, add gold, and ask what happens when the quadrant turns against everything at once. Then we test whether a price-only trend rule, with no knowledge of the economy, solves for that blind spot.

Four regimes, 120 years

The equity sleeve holds sixteen countries in equal weight. The bond sleeve does the same with long government bonds. Gold is the third sleeve. The baseline portfolio holds a third of each and rebalances every year. All returns are in real US dollars, unhedged, deflated by US CPI. The reference investor is a dollar-based global portfolio holder.

Classifying each year 1901–2020 by US real GDP growth (from the Maddison Project) and US inflation against their own trailing ten-year averages produces four roughly equally populated quadrants, with twelve neutral years sitting within one percentage point of trend on both axes.

The benign quadrants work as advertised. Equities compound at 9.2% real when growth is rising and inflation falling. Bonds compound at 7.2% when both are falling. That is the deflation rally. The equal-weight baseline returns between 2.8% and 4.2% in three of the four cells.

The fourth quadrant is the reason for this article.

The quadrant nothing survives

When growth falls and inflation rises together (stagflation), equities lose 4.8% real per year, bonds lose 2.9%, and gold does essentially nothing (0.1%). Bills also lose because the short rate is held below inflation. The equal-weight baseline compounds at minus 5% real per year, and no fixed combination of the three sleeves turns that positive. Every component is negative.

One year in five sits in this quadrant. It is not a tail event. It is a recurring structural feature of developed-market returns.

Gold, after the peg

For most of the 1901–2020 sample, gold's price was fixed by statute. Sorting by the end of the Bretton Woods system in 1971 makes the split visible. Before the float, gold lost most when inflation rose, because a fixed nominal price under rising prices is a loss by construction. After 1972, gold compounded at 22% a year in the years inflation rose and lost 6% in the years it fell (source: MeasuringWorth annual London gold price). That is the behaviour a store of value should show, and the 1971 boundary is why it was invisible across the full sample.

The evidence is 49 years of a single asset with 22% annual volatility, and the inflation-up cell holds just nine of those years. It is enough to keep gold in the portfolio on its own merits, but not enough to rely on it as a stand-alone stagflation hedge.

The price-only rule that doesn't read the economy

The solution proposed by the source research is not a more sophisticated asset mix. It is the opposite: a rule that holds less when recent returns say to.

The simplified rule: take the prior calendar year's total return for each sleeve, divide by its trailing ten-year volatility, clip the result to the range [-1, +1], and set that sleeve's weight to one-third times (0.5 + 0.5 × signal). A strong year keeps the sleeve at its full third. A flat year halves it. A bad year removes it. There is no leverage, no shorting, and the multiplier for year t is set from data ending December of year t−1.

The rule sees prices only. It has no growth input, no inflation input, and no knowledge of which quadrant it is in.

What it did

Across 120 years, the rule adds a small amount of return (0.3 points annually) while removing a third of the volatility. Maximum drawdown falls from 59% to 17%. The longest stretch below a previous high contracts from 21 years to eight. Average exposure is 55%. The portfolio sits in cash close to half the time, which is what the drawdown reduction costs.

The years in which the rule held nothing at all (1908, 1919–1921, 1947–1949) were every one a year in which all three sleeves lost real value together. The rule found the common shock from price alone.

By regime

The claim to test: does a price-only rule reduce exposure in the exact quadrant where exposure is harmful, without being told which quadrant it is in?

It does. Bond exposure falls from four-fifths of its full weight in the benign quadrant to two-fifths in stagflation. In the stagflation cell the rule compounds at +0.8% real against −5.0% for the baseline. The rule does not read the regime. It reads last year's loss, and the regime is what caused it.

Insurance has a premium

A rule that improved every regime would be fitted, not found. This one gives back return in the two quadrants where inflation is falling. The deflation quadrant costs most: when growth and inflation both fall, bonds rally hard and the recovery is fast. The rule has cut bond exposure after the preceding bad year and re-enters late. The gap in that quadrant is 2.4 percentage points a year across 28 years, the largest single cost in the sample.

Over the 1982–2020 bull market, the rule trailed the baseline by 1.4 points a year. That is the insurance premium, paid in the years when nothing goes wrong.

Whether the 1914–1945 and 1946–1967 columns (both of which the rule handled better than the baseline) are worth it depends on the reader's horizon, not on the data.

What this test cannot settle

Few active traders would use an annual signal, and the annual simplification here collapses four lookbacks to one, removes leverage and shorting, and charges no financing costs (all in the conservative direction). The daily version from the Beyond Passive Investing series (four lookbacks of 21, 63, 126 and 252 trading days, vol-scaled, with financing) showed the same shape from 1968: about half the drawdown for roughly flat return, with the value-add concentrated in the inflationary regimes.

The sixteen historical countries cannot be bought as such. Eight now share a currency. The implementable unit is the political zone: about ten of these in the developed world, five in Europe, and equal weight across zones (not across countries) is the target. That portfolio exists as ETFs with prices back to the early 2000s. The annual test can settle whether the mechanism works. It cannot replace a real-cost, real-ticker implementation.

Where this meets our work

The insight here (that "the missing asset" is not an instrument but permission to hold less) is the same principle that drives our portfolio-level rebalance trigger at RiskHarvest. Rather than forecast which regime comes next, we pre-commit to a rule that reduces exposure when the momentum signal says to. The trend rule tested here is an extreme simplification of that logic: no volatility target, no leverage, no cross-sectional ranking, just a sleeve-by-sleeve go/no-go based on trailing returns.

Our vol-targeted risk sizing applies this same "permission to hold less" mechanic at a finer grain. Rather than cut a third of the portfolio to cash, we scale each position continuously toward its risk budget. The result is the same structural hedge (reduced exposure during the regimes that hurt most) without the blunt on/off character of the annual trend rule. The mechanism generalises: the most reliable hedge is not a better asset, but a rule that lets you hold less of the assets that are currently falling.

One key message

Over 120 years and sixteen countries, the regime that destroys every conventional asset simultaneously is not a tail event. It is a recurring structural feature of developed-market returns. No fixed allocation solves it. A price-only trend rule, blind to the economy, cuts exposure in exactly that quadrant without needing to forecast it. The cost is a modest drag during benign regimes. Whether that trade-off is worth it depends on whether you think the next twenty years will look more like 1914–1967 or 1982–2020.

Data sources: Jordà-Schularick-Taylor Macrohistory Database, Release 6; MeasuringWorth annual London gold price; Maddison Project real GDP per capita. All rates are compound annual growth rates in real US dollars. Sleeves are equal-weighted across sixteen countries, unhedged, deflated by US CPI. Years 1901–2020. See NBER and macrohistory.net for the underlying data.

This article is educational and does not constitute investment advice. Past performance is not indicative of future results.