Uncorrelated Return Streams

Fifteen good, uncorrelated streams

Ray Dalio states his central portfolio-construction rule as a mantra: "15 good, uncorrelated return streams, because if they're engineered to have about the same expected return and you have that kind of diversification, you will lower the risk by about 80%, which raises the return-to-risk ratio by a factor of five."¹ The claim he makes is specific. It is not that diversification is generally prudent, but that a particular number of genuinely independent streams, tuned to comparable expected returns, produces a measurable reduction in volatility while leaving expected return intact.

The mechanism Dalio describes rests on correlation. When holdings move together, their volatility stacks; when they move independently, much of it cancels. In his framing, a set of uncorrelated streams of equal expected return and equal volatility produces a portfolio whose risk falls roughly as one over the square root of the number of streams, so that around 15 streams yields something close to the 75 to 80 percent risk reduction he rounds to. The improved return-to-risk ratio then gives an operator a choice: hold the same return while taking far less risk, or apply leverage to reach the original risk level and capture the higher risk-adjusted return. Dalio says Bridgewater uses the second version, levering the diversified portfolio up to the desired risk level, and he presents this as the structural basis of the firm's pure-alpha approach.

Fifteen as a practical threshold

Dalio is careful that 15 is not a magic number but a practical threshold past which additional streams add little further risk reduction. He names the binding constraints himself. True uncorrelation is rare, because most apparent diversifiers turn out to be correlated to equity risk during market stress. Each stream must carry a positive expected return after costs, since mixing good ideas with bad ones does not help. And the streams must have about the same expected return per unit of risk, or the portfolio concentrates in whichever stream is most volatile. He frames the work as engineering rather than asset selection, describing streams that are built to be genuinely different sources of return rather than repackaged versions of the same underlying risk. For Dalio the portfolio is a designed object, and its structure, rather than any individual holding in it, is the primary instrument.

The 1982 origin

Dalio traces the insight to a failure rather than a success. In 1982 he had read the Latin American debt crisis correctly and taken a concentrated position anticipating a severe contraction. The Federal Reserve eased, the stock market rallied, and he was wiped out. His stated lesson was not to become more accurate about macro forecasting but to stop betting concentrated no matter how right a position feels, because even a strong analyst is wrong about market direction often enough that a concentrated portfolio will eventually fail. In his telling, diversification is not a way of hedging ignorance but an engineering response to the unavoidable distribution of outcomes.

He points to Bridgewater's long record, reported at roughly 11.8 percent annual return since 1982 with most down years around negative 2 percent and a worst year of about negative 13 percent in the 2020 pandemic, as his evidence that the framework holds up in practice. He also offers a simplified retail version, since he does not expect an individual to build 15 streams from scratch: start with inflation-protected securities, add 10 to 15 percent in gold as a stress-period diversifier, and spread exposure across geographies and asset classes. The all-weather strategy, which weights assets by their risk contribution across economic regimes, is in his account the institutionalized form of the same principle. The framework's own hard part, which Dalio does not minimize, is that engineering true uncorrelation is difficult, and streams that look independent in calm markets can reveal hidden correlations precisely when the diversification is most needed.