How Combining Two Risky Assets Can Beat Owning Either Alone
An investor holding stocks and bonds separately often assumes the combined portfolio's risk is simply an average of the two pieces' individual risk. The actual math says something more useful: blending two assets that do not move in lockstep can produce a combined risk lower than either piece carries on its own.
Why correlation is the hidden variable
The variance of a two-asset portfolio depends on three things: how risky each asset is on its own, and, critically, how the two move together, captured by their correlation, a number between -1 and 1. The full formula is: portfolio variance = (w1² × σ1²) + (w2² × σ2²) + (2 × w1 × w2 × ρ × σ1 × σ2), where w1 and w2 are the portfolio weights, σ1 and σ2 are the individual standard deviations, and ρ is the correlation between the two assets' returns.
The first two terms are unavoidable, each asset contributes risk in proportion to the square of its own weight and volatility. The third term is where the interesting behavior lives: it grows smaller, and can even turn negative, as ρ falls. At ρ = 1, the two assets always move together and the formula collapses to a simple weighted average of the two standard deviations, no diversification benefit at all. At ρ below 1, some of each asset's individual volatility gets cancelled by the other's offsetting movement, and portfolio risk falls below that simple weighted average. At ρ below zero, the effect strengthens further, and it becomes mathematically possible for the combined portfolio to carry less risk than either individual asset alone. This last result surprises many investors on first encounter, since intuition suggests that adding a more volatile asset to a portfolio should always raise risk, and the formula's third term is precisely the mechanism that can overturn that intuition.
Two worked examples
Consider Stock A, with an expected return of 8% and a standard deviation of 12%, and Stock B, with a higher expected return of 14% but a higher standard deviation of 22%, with a correlation of 0.2 between them, a fairly typical value for two stocks in different industries. A simple 50/50 blend has an expected return of exactly 0.5 × 8% + 0.5 × 14% = 11%. Its variance is: (0.5² × 0.0144) + (0.5² × 0.0484) + (2 × 0.5 × 0.5 × 0.2 × 0.12 × 0.22) = 0.0036 + 0.0121 + 0.00264 = 0.01834, giving a standard deviation of √0.01834 ≈ 13.54%. Compare that to a naive weighted average of the two standard deviations, 0.5 × 12% + 0.5 × 22% = 17%: the actual portfolio risk, 13.54%, sits more than 3 percentage points below that naive figure, purely because the two assets are not perfectly correlated.
The second, more striking result comes from solving for the specific weighting that minimizes portfolio variance entirely, given by w1* = (σ2² - ρσ1σ2) / (σ1² + σ2² - 2ρσ1σ2). Plugging in the same two assets: numerator = 0.0484 - (0.2 × 0.12 × 0.22) = 0.0484 - 0.00528 = 0.04312; denominator = 0.0144 + 0.0484 - (2 × 0.00528) = 0.0628 - 0.01056 = 0.05224. That gives w1* = 0.04312 / 0.05224 ≈ 0.825, so roughly 82.5% in Stock A and 17.5% in Stock B minimizes risk. Working through the variance formula at those weights gives a portfolio standard deviation of approximately 11.3%, actually lower than Stock A's own 12% standard deviation, even though the portfolio has added a considerably more volatile second asset to the mix, and its expected return has risen slightly to about 9.05% in the process, since Stock B's higher expected return still gets a small weight.
A third case: what negative correlation changes
The two worked examples above both used a modestly positive correlation of 0.2. It is worth seeing explicitly what changes when correlation turns negative, since that is the more dramatic, and rarer, version of the diversification benefit. Keep Stock A (8% expected return, 12% standard deviation) and Stock B (14% expected return, 22% standard deviation) from before, but now assume a correlation of -0.3 instead of 0.2, plausible for two assets with genuinely different, sometimes opposing, return drivers.
The minimum-variance weight formula becomes: numerator = 0.0484 - (-0.3 × 0.12 × 0.22) = 0.0484 + 0.00792 = 0.05632; denominator = 0.0144 + 0.0484 - (2 × -0.00792) = 0.0628 + 0.01584 = 0.07864. That gives w1* = 0.05632 / 0.07864 ≈ 0.716, so about 71.6% in Stock A and 28.4% in Stock B. Computing the resulting variance: w1²σ1² = 0.5127 × 0.0144 = 0.007383; w2²σ2² = 0.0807 × 0.0484 = 0.003906; the cross term is 2 × 0.716 × 0.284 × (-0.00792) = -0.003217. Summing: 0.007383 + 0.003906 - 0.003217 = 0.008072, a standard deviation of √0.008072 ≈ 8.98%, meaningfully below the 11.3% achieved earlier with correlation of 0.2, and well below Stock A's own 12% standard deviation. The negative correlation term is now subtracting from total variance rather than merely adding less than a perfectly correlated pair would, which is why negative correlation produces a noticeably larger risk-reduction benefit than a low positive correlation does, even though both directionally help.
What stock and bond correlation history shows
The most widely used real-world application of this math is the classic stock-bond blend, and its historical correlation record is instructive precisely because it has not been constant. Across much of the period from the late 1990s through the late 2010s, U.S. stocks and high-quality government bonds displayed a correlation that was frequently low or outright negative, meaning bonds tended to hold up or even rally during sharp stock selloffs, exactly the pattern that makes the two-asset math above so attractive for smoothing a portfolio's ride.
That pattern is not a law of nature. In 2022, stocks and bonds fell together, a rare and uncomfortable break from the more typical negative-correlation pattern, driven by a shared sensitivity to a sharp, unexpected rise in interest rates that hurt bond prices directly and hurt stock valuations by raising the discount rate applied to future earnings at the same time. Investors who had come to expect bonds to reliably cushion a stock decline learned, in a single painful year, that the correlation input to this formula is an estimate drawn from history, not a permanent guarantee, and it can shift when a common macro factor, interest rates in that case, dominates both asset classes simultaneously.
Using this in a real portfolio
The practical takeaway is not to chase the single pair of assets with the lowest historical correlation and load up on it uncritically; extremely low correlation often comes paired with a lower expected return, and a portfolio optimized purely for minimum variance can end up with a return too low to meet a long-term goal. The more useful frame is choosing a blend that meaningfully improves the risk-return tradeoff versus either asset alone, without sacrificing so much expected return that the portfolio no longer serves its purpose. A 70/30 or 80/20 tilt toward the higher-expected-return asset, informed by but not slavishly following the pure minimum-variance weight, is a common practical compromise.
For an investor building wealth through a working career, the classic building blocks are a broad equity index and a high-quality bond index, but the same math applies to any second, genuinely differentiated risky asset: real estate, a diversified international equity sleeve, or, for a business owner or partner in a professional practice, business equity that behaves differently from the broader public market. What matters is the actual correlation, estimated honestly over a long enough period to include at least one stress event, not just the label of the asset class.
For a high-earning professional, this second risky asset is often, in practice, an ownership stake in a private business or practice rather than a publicly traded fund, and the same formula still applies conceptually even though the inputs are harder to estimate precisely. A dentist who owns both a diversified public equity portfolio and an equity stake in the practice is holding a two-asset portfolio whether or not it is ever formally analyzed as one, and the honest question worth asking is how correlated the practice's value is likely to be with the broader stock market during a downturn, since a local economic shock that hits patient volume can coincide with, rather than offset, a broader market decline, unlike the more classically low-correlation stock-bond pairing. Recognizing this honestly, rather than assuming any second asset automatically diversifies the first, is the difference between a portfolio that actually behaves the way the formula predicts and one that only looks diversified on paper until the correlation that mattered most finally shows up, usually during precisely the stretch of time when the investor can least afford the surprise, which is the whole reason this kind of checking is worth doing before a crisis, not after one.
Actionable breakdown
- Estimate correlation honestly before combining two assets.
- Use a long enough history to include at least one real crisis period.
- Do not assume a low calm-market correlation will hold under stress.
- Compute actual portfolio variance rather than eyeballing an average.
- Apply the full two-asset formula, not a simple weighted average of risks.
- Consider the minimum-variance weight as a reference point, not a target.
- Balance it against the return you actually need to meet your goal.
- Revisit the correlation assumption periodically.
- Macro regime shifts, like the 2022 rate spike, can move it materially.
Common pitfalls
The most common pitfall is assuming any two different-looking assets automatically reduce risk when combined. Correlation, not superficial variety, is what does the work, and two assets that look different on the surface can still move together closely enough to provide little real benefit.
A second pitfall is overlooking that correlations often rise sharply during exactly the market stress periods when diversification matters most, a pattern that shows up across asset classes, not just within stocks.
A third pitfall is chasing minimum variance too aggressively, accepting a meaningfully lower expected return in exchange for a modest additional risk reduction, when a small tilt toward the higher-returning asset, informed by but not fully constrained by the minimum-variance weight, often serves a long-term goal better. A fourth pitfall is applying the formula with correlation and volatility inputs estimated from a very short recent window, a year or two of data can produce a correlation estimate that swings widely and unreliably compared to an estimate drawn from a full market cycle spanning at least one meaningful downturn.
The bottom line
Combining two assets with a correlation meaningfully below 1 can lower total portfolio risk below what either asset carries on its own, but the benefit depends on an honestly estimated correlation that is checked against real crisis periods, not just calm-market history.
Related reading: building an asset allocation, how bonds work, real estate and REITs, how diversification reduces risk, the capital allocation line.