Splitting a Portfolio Across Stocks, Bonds, and Bills
Most investors treat their stock-bond split as a single dial turned by gut feeling, which leaves real diversification benefit on the table. Three assets, equities, bonds, and a risk-free instrument like Treasury bills, can be combined in a specific order that produces a strictly better risk-return tradeoff than any ad hoc mix.
The two step decision hiding inside a three asset portfolio
A portfolio built from stocks, bonds, and Treasury bills looks like a single three-way allocation problem, but it is really two separate decisions stacked on top of each other. The first decision is which combination of stocks and bonds, the two risky pieces, produces the best possible risk-adjusted return on its own, independent of how much of the total portfolio will ultimately sit in either asset. The second decision is how much of the total portfolio to put into that stock and bond mix versus how much to hold in bills, the risk-free piece.
This separation matters because it implies every investor, regardless of how conservative or aggressive they are, should in principle hold the same stock-to-bond ratio within their risky sleeve. A retiree and a twenty five year old disagree about how much risk to carry overall, and that disagreement should show up as a different weight on bills, not as a different ratio of stocks to bonds inside the risky portion. Few portfolios in practice are built this cleanly, but the framework still tells you what you give up when you deviate from it, which is usually a lower return for the same risk taken, or more risk for the same return.
The math: finding the best stock and bond mix
The stock and bond combination that maximizes reward per unit of risk, sometimes called the tangency portfolio, depends on each asset's expected excess return over the risk-free rate, each asset's variance, and the covariance between the two. The weight to place in stocks, with the remainder in bonds, is w_S = [(E(R_S)-R_f) × σ_B² - (E(R_B)-R_f) × Cov(S,B)] / [(E(R_S)-R_f) × σ_B² + (E(R_B)-R_f) × σ_S² - ((E(R_S)-R_f)+(E(R_B)-R_f)) × Cov(S,B)].
Suppose stocks have an expected return of 10% and a standard deviation of 20%, bonds have an expected return of 5% and a standard deviation of 8%, the correlation between them is a modest 0.15, and bills yield 3%. The covariance is 0.15 × 20% × 8% = 0.0024. Excess returns are 7% for stocks and 2% for bonds. Plugging into the formula, the numerator is 0.07 × 0.0064 - 0.02 × 0.0024 = 0.000448 - 0.000048 = 0.0004, and the denominator is 0.07 × 0.0064 + 0.02 × 0.04 - 0.09 × 0.0024 = 0.000448 + 0.0008 - 0.000216 = 0.001032. That gives w_S = 0.0004 / 0.001032 = 38.8%, so bonds get the remaining 61.2%.
This mix has an expected return of 0.388 × 10% + 0.612 × 5% = 3.88% + 3.06% = 6.94% and a variance of 0.388² × 0.04 + 0.612² × 0.0064 + 2 × 0.388 × 0.612 × 0.0024 = 0.00602 + 0.00240 + 0.00114 = 0.00956, giving a standard deviation of 9.78%. Its Sharpe ratio, (6.94% - 3%) / 9.78% = 0.40, beats stocks alone at 7% / 20% = 0.35 and bonds alone at 2% / 8% = 0.25. Neither asset held by itself matches what the blend achieves, purely because the two return streams do not move in lockstep.
It is worth seeing how sensitive that 38.8% figure is to the underlying assumptions, since real investors rarely have certainty about expected returns. Lower the assumed bond return from 5% to 4%, holding everything else fixed, and the excess return on bonds falls from 2% to 1%. Recomputing, the numerator becomes 0.07 × 0.0064 - 0.01 × 0.0024 = 0.000448 - 0.000024 = 0.000424, and the denominator becomes 0.07 × 0.0064 + 0.01 × 0.04 - 0.08 × 0.0024 = 0.000448 + 0.0004 - 0.000192 = 0.000656, giving w_S = 0.000424 / 0.000656 = 64.6%. A one-point change in the assumed bond return, small by the standards of typical forecasting error, shifted the optimal stock weight from 38.8% up to 64.6%. This kind of sensitivity is exactly why the weight should be treated as a reasoned estimate updated periodically, not a number computed once and trusted indefinitely.
Adding bills: a second worked example
Once the tangency mix of stocks and bonds is set, the second decision is how much weight, y, to place in that mix versus bills, with (1-y) in bills. This traces out a straight line on a risk-return chart running from the risk-free point through the tangency portfolio, and any point on that line dominates any point achievable by mixing stocks and bonds directly without bills.
Take an investor targeting a portfolio standard deviation of 6%, lower than the 9.78% carried by the tangency mix itself. Solving y × 9.78% = 6% gives y = 61.4%, so 61.4% of the total portfolio sits in the tangency mix and 38.6% sits in bills. The expected return is 3% + 0.614 × (6.94% - 3%) = 3% + 0.614 × 3.94% = 3% + 2.42% = 5.42%. Translating that back into the underlying three-asset weights, stocks get 0.614 × 38.8% = 23.8%, bonds get 0.614 × 61.2% = 37.6%, and bills get the remaining 38.6%.
A more aggressive investor targeting 15% total risk instead sets y = 15% / 9.78% = 153%, meaning they borrow an amount equal to 53% of their own capital to hold more than 100% in the tangency mix. Their expected return becomes 3% + 1.53 × 3.94% = 3% + 6.03% = 9.03%, at the cost of real borrowing rates and margin risk that the idealized formula ignores entirely. Both investors hold stocks and bonds in the identical 38.8 to 61.2 ratio; only their exposure to that combined position differs.
What market history shows about the stock bond mix
Realized correlations between broad U.S. stocks and investment grade bonds have not been stable over time, which is the main practical complication with this framework. Across much of the 2000s and 2010s, stock and bond returns were frequently negatively correlated, meaning bonds tended to rally when stocks fell, which made the blended portfolio's realized risk reduction even better than a simple positive-correlation assumption would predict. In 2022, that relationship inverted sharply: both stocks and long-duration bonds fell together as interest rates rose quickly, and the correlation between them turned meaningfully positive for an extended stretch, the opposite of what several prior decades of data had trained investors to expect.
This history is a caution against treating any single correlation estimate, including the 0.15 used above, as a fixed constant. A tangency weight computed from trailing five-year data gathered during a low-rate, negative-correlation period can look quite different from one computed using data spanning a rising-rate, positive-correlation period. The direction of the effect on the optimal weight is not fixed either: rising bond correlation with stocks reduces the diversification benefit bonds provide, which typically lowers bonds' share of the optimal risky mix rather than raising it.
Bond maturity, often called duration, adds a second layer of history worth knowing before treating "bonds" as a single homogeneous asset. Long-duration Treasury bonds carry standard deviations closer to 12% to 15% in typical conditions, well above the 8% used in the examples here, which is closer to what a diversified intermediate-term investment grade bond fund has historically shown. Substituting long-duration bonds into the formula above, holding return assumptions fixed, would lower the tangency weight on bonds simply because their higher volatility makes them a less efficient risk-reducer per unit of expected return, independent of any change in correlation. The specific bond exposure chosen, not just "stocks versus bonds" as a label, materially changes the arithmetic.
Using this in a real portfolio
Few individual investors sit down and solve the tangency formula by hand, and that is fine: target-date funds, balanced funds, and most robo-advisor portfolios are built on some version of this same two-step logic, using far more sophisticated return and covariance estimates than a single correlation number pulled from one example. What the framework offers a self-directed investor is a way to separate two questions that otherwise get muddled together. The first is whether your stock-to-bond ratio within the risky sleeve is actually well diversified given current correlation and volatility conditions, rather than an arbitrary 60/40 inherited from habit. The second is whether your overall exposure to that mix, versus cash or bills, matches your actual capacity and willingness to withstand a drawdown.
A useful diagnostic is to compute your portfolio's implied Sharpe ratio and compare it against the Sharpe ratio of the tangency mix under reasonable long-run assumptions. A portfolio sitting meaningfully below the achievable frontier, holding, say, an 80% stock and 20% bond split when the tangency analysis suggests something closer to 40/60 for the risky sleeve itself, is not necessarily wrong, but the investor should understand they are choosing a specific point off the efficient line, whether out of a preference for simplicity, a belief the inputs are mismeasured, or plain inertia.
Actionable breakdown
- Solve the risky mix and the bills weight separately.
- Find the stock-to-bond ratio that maximizes Sharpe ratio first.
- Only then decide how much of that mix to hold versus bills.
- Treat correlation estimates as a range, not a fixed input.
- Recompute the tangency mix under both low and high correlation cases.
- Notice how much the optimal bond weight shifts between them.
- Use Sharpe ratio, not headline return, to judge the blend.
- A mix with a lower expected return can still be the better choice.
- Compare the blend's Sharpe ratio to each asset held alone.
- Be honest about leverage if you push y above 100%.
- Use your real borrowing rate, not the risk-free rate, in any projection.
- Account for margin calls during sharp drawdowns.
Common pitfalls
The most common pitfall is assuming a fixed stock-to-bond correlation, often borrowed from a decade with unusually favorable diversification conditions, and treating the resulting optimal weight as permanent. The 2022 breakdown in the classic negative-correlation pattern is the clearest recent illustration of why this assumption needs periodic revisiting rather than set-and-forget treatment.
A second pitfall is conflating the risky-mix decision with the overall risk decision, leading investors to think that raising their bond allocation is always the way to reduce risk. Sometimes the more efficient move is to leave the risky mix's own ratio unchanged and instead shift weight toward bills, which reduces risk without sacrificing the mix's internal diversification benefit.
A third pitfall is leaning on short-window historical averages, sometimes just one or two years, to estimate expected returns and volatilities, which produces tangency weights that swing wildly from one calculation to the next. Longer estimation windows, or blending historical data with forward-looking valuation-based estimates, produce far more stable and usable weights.
The bottom line
Find the stock and bond ratio that maximizes reward per unit of risk first, then use bills to scale your overall exposure to that mix up or down to match your own risk tolerance.
Related reading: building an asset allocation, how bonds behave in a portfolio, what actually counts as risk-free, the capital allocation line, extending this to many assets with the Markowitz model.