01 · Correlation regimes

Correlation changes with the regime.

Long-run evidence shows that stock–bond correlation has moved between positive and negative regimes. A negative estimate from one period is therefore not a contractual hedge for the next drawdown.

Rolling 20-year correlation of annual S&P 500 and 10-year US Treasury total returns from 1947 to 2025.Enlarge chart
Rolling 20-year Pearson correlation of annual nominal total returns. The first estimate uses 1928–1947; the final estimate uses 2006–2025. Long windows reveal structural change, but can smooth shorter crisis dynamics.

Research chart

Scroll horizontally to inspect the full chart. The chart area also supports arrow keys.

Rolling 20-year correlation of annual S&P 500 and 10-year US Treasury total returns from 1947 to 2025.

Rolling 20-year Pearson correlation of annual nominal total returns. The first estimate uses 1928–1947; the final estimate uses 2006–2025. Long windows reveal structural change, but can smooth shorter crisis dynamics.

↘ / ↗

Growth shock

Disinflation and policy easing can support duration while equities weaken—allowing the traditional offset to work.

↘ / ↘

Inflation shock

Higher inflation, policy rates or term premia can pressure equities and nominal bonds at the same time.

  • RegimesStock–bond correlation is time-varying and has changed sign across long samples.
  • DirectionalitySystematic trend strategies can change direction, but only after a trend becomes observable.
  • ComplementarityResilience comes from different return mechanisms—not more labels for the same macro exposure.
02 · Stress regimes in the data

Different shocks. Different outcomes.

The contrast between 2008 and 2022 is the central visual lesson: duration provided a powerful offset during a growth and liquidity shock, but not during an inflation and policy shock.

Annual total returns for the S&P 500, 10-year US Treasury and a 60/40 illustration in 2002, 2008 and 2022.Enlarge chart
The 60/40 illustration combines 60% S&P 500 and 40% 10-year US Treasury annual returns and assumes annual rebalancing, before fees, taxes and transaction costs. It is not an investable portfolio or Tensor performance.

Research chart

Scroll horizontally to inspect the full chart. The chart area also supports arrow keys.

Annual total returns for the S&P 500, 10-year US Treasury and a 60/40 illustration in 2002, 2008 and 2022.

The 60/40 illustration combines 60% S&P 500 and 40% 10-year US Treasury annual returns and assumes annual rebalancing, before fees, taxes and transaction costs. It is not an investable portfolio or Tensor performance.

Scatter plot of annual S&P 500 and 10-year US Treasury returns, coloured by inflation regime, from 1928 to 2025.Enlarge chart
Each point is one calendar year. Red indicates US inflation at or above 3%; teal indicates inflation below 3%. The lower-left quadrant contains years in which both assets lost money.

Research chart

Scroll horizontally to inspect the full chart. The chart area also supports arrow keys.

Scatter plot of annual S&P 500 and 10-year US Treasury returns, coloured by inflation regime, from 1928 to 2025.

Each point is one calendar year. Red indicates US inflation at or above 3%; teal indicates inflation below 3%. The lower-left quadrant contains years in which both assets lost money.

Data: Aswath Damodaran, Historical Returns on Stocks, Bonds and Bills, 1928–2025. The S&P 500 series includes dividends; Treasury returns use a constant-maturity 10-year repricing methodology based on FRED yields. Tensor calculations. Historical results are not predictive.

03 · Trend following

Trend following is not a static hedge.

A generic time-series momentum implementation can hold long exposure after positive trends and short exposure after negative trends across equity indices, rates, currencies and commodities. That directional flexibility can create a return path unlike long-only assets when moves persist.

Why the mechanism can diversify

  • Direction can change as price trends change.
  • Risk can be distributed across multiple asset classes.
  • Persistent macro repricing can create time for adaptation.

Why the mechanism can disappoint

  • A sudden shock can occur before the signal turns.
  • Sharp reversals can create whipsaw losses.
  • Trading costs, contract rolls and implementation matter materially.
Historical research: evidence and limitations
58

liquid futures

Studied across equity index, currency, commodity and bond futures by Moskowitz, Ooi & Pedersen.

1–12

month persistence

Intermediate-horizon return persistence reported in the study.

1880

historical start

Hurst, Ooi & Pedersen extend trend-following evidence across global markets to 1880.

8 / 10

largest 60/40 crises

Their constructed strategy was positive in eight of ten largest crisis periods studied.

Interpretation discipline. “Eight out of ten” is historical evidence from a constructed research strategy—not a promise that any CTA, implementation or future crisis will be profitable.

Allocator implications

Test the sources of diversification.

Two engines should be combined because their economic behaviour is expected to differ—not because two backtests make a chart look smoother. The test is whether complementarity survives costs, common-factor decomposition, crisis analysis and out-of-sample evaluation.

01

Stress multiple regimes

Use rolling and crisis-window correlations rather than one full-sample covariance matrix.

02

Test common exposures

Measure equity, duration, USD, commodity, volatility and momentum dependence.

03

Preserve the path

Stress gaps, persistent trends, policy reversals and liquidity assumptions—not only end-point returns.

04

Model implementation drag

Include turnover, bid–ask, slippage, financing, rolls and stressed costs.

05

Publish failure modes

Show where each engine loses, where both lose together and how quickly diversification can disappear.

06

Keep evidence auditable

Document data lineage, regime definitions, holdout choices and human approval.

The objective is not permanent protection. It is fewer ways for one regime to dominate the entire portfolio.
Discuss our research architectureExplore the strategy concepts