ABC Tools

The Efficient Frontier and the Rebalancing Bonus

After reading this you will be able to build a Markowitz efficient frontier by hand for a few assets, find the minimum-variance and maximum-Sharpe portfolios, and estimate the extra return that regular rebalancing can harvest from diversification.

What the frontier is, with one hook

Suppose you can hold three assets: stocks (expected return 7%, volatility 16%), bonds (3% and 6%), and gold (4% and 15%). Mix them in every possible proportion and each mix lands somewhere on a risk-return map. Most mixes are wasteful: for the same risk you could have earned more, or for the same return you could have taken less risk. The efficient frontier is the upper-left edge of that cloud, the set of mixes that give the most return for each level of risk.

Here is the surprising part. A 60/40 stock-bond mix in this example has a volatility of roughly 10%, which is lower than the 6% of bonds only in the sense that... no, it sits between the two. But because stocks and bonds move a little against each other (correlation -0.1), the blend is calmer than a naive average would predict. That gap is the whole point. Low correlation lets a portfolio be steadier than the weighted average of its parts, and that steadiness turns into extra compounded return over time.

When to use it, and when not

Use the frontier to compare a handful of asset classes when you already have views on their expected returns, volatilities and how they move together. It is a thinking tool for allocation: it shows you the shape of the trade-off and where the sensible corners are.

Do not treat its output as a precise instruction. The frontier is only as good as its inputs, and expected returns are the weakest input of all. Nobody knows next year's stock return. A change of one percentage point in an expected return can swing the optimal weights by twenty points or more. Treat the frontier as a map of possibilities, not a GPS route.

Optimizers love to concentrate. If you feed in a slightly higher return for one asset, the math will happily pour most of the portfolio into it. Always check whether the "optimal" weights are stable when you nudge the inputs. If they are not, trust a rounder allocation over the razor-sharp one.

The math, from the ground up

A portfolio is a set of weights w_i that sum to 1. Its expected return is just the weighted average of the parts:

\mu_p = \sum_i w_i \mu_i

Here \mu_i is the expected return of asset i and w_i its weight. Return is linear, so nothing clever happens here.

Risk is where diversification lives. The portfolio variance combines each asset's own variance with every pair's covariance:

\sigma_p^2 = \sum_i \sum_j w_i w_j \sigma_i \sigma_j \rho_{ij}

In this formula \sigma_i is asset i's volatility, and \rho_{ij} is the correlation between assets i and j. When i = j the correlation is 1, so the diagonal terms are the plain variances w_i^2 \sigma_i^2. The off-diagonal terms carry the correlation, and when a correlation is low or negative those terms shrink the total. Portfolio volatility is \sigma_p = \sqrt{\sigma_p^2}.

The Sharpe ratio measures return earned per unit of risk, above what a risk-free asset pays:

S = \frac{\mu_p - r_f}{\sigma_p}

Here r_f is the risk-free rate. The frontier point with the highest Sharpe ratio is the tangency portfolio, the best trade of excess return for risk. The point with the smallest \sigma_p is the minimum-variance portfolio.

Worked example with the demo data

Three assets, a risk-free rate of 2%

Load the demo: Stocks 7 16, Bonds 3 6, Gold 4 15, with correlations Stocks-Bonds -0.1, Stocks-Gold 0.1, Bonds-Gold 0.15, and r_f = 2\%. Work returns and vols in decimals: stocks (0.07, 0.16), bonds (0.03, 0.06), gold (0.04, 0.15).

  1. Take a candidate mix of 50% stocks, 40% bonds, 10% gold. Expected return is 0.5(0.07) + 0.4(0.03) + 0.1(0.04) = 0.051, or 5.1%.
  2. Diagonal variance terms: 0.5^2(0.16)^2 = 0.006400, 0.4^2(0.06)^2 = 0.000576, 0.1^2(0.15)^2 = 0.000225.
  3. Cross terms (each counted twice). Stocks-bonds: 2(0.5)(0.4)(0.16)(0.06)(-0.1) = -0.000384. Stocks-gold: 2(0.5)(0.1)(0.16)(0.15)(0.1) = 0.000240. Bonds-gold: 2(0.4)(0.1)(0.06)(0.15)(0.15) = 0.0001080.
  4. Sum: 0.006400 + 0.000576 + 0.000225 - 0.000384 + 0.000240 + 0.000108 = 0.007165. So \sigma_p = \sqrt{0.007165} \approx 0.0847, about 8.5%.
  5. Sharpe: (0.051 - 0.02) / 0.0847 ≈ 0.366.

Search across all long-only mixes and two corners stand out. The minimum-variance portfolio leans heavily on bonds (roughly 80% bonds, 15% stocks, 5% gold) with a volatility near 5.6%. The maximum-Sharpe portfolio holds more stocks (around 45% stocks, 50% bonds, 5% gold) with a Sharpe close to 0.40. The exact weights shift with tiny input changes, so read them as neighborhoods, not decimals.

Each dot is a random mix. The upper-left edge is the efficient frontier; anything below it is dominated.

Reading and interpreting the results

Three features carry the meaning. The cloud shows how much waste is possible: most random mixes sit well below the edge. The frontier line is the menu you should actually choose from. The two corner points anchor your choice.

Minimum-variance portfolio
The calmest mix you can build. Pick it if you care about stability above all and have no strong return view.
Maximum-Sharpe (tangency) portfolio
The best reward-to-risk trade. If you can also hold cash at r_f, every efficient portfolio is a blend of this one and cash, so it is the natural risky anchor.
Risk-free rate
Raising r_f tilts the tangency portfolio toward higher-return, higher-risk assets, because the bar for "worth the risk" rises with it.

Between the two corners the frontier is nearly flat: you can move from minimum-variance toward maximum-Sharpe and pick up meaningful return for modest extra risk. Past the tangency point the line curves sharply, and each extra unit of return costs a lot more volatility.

The rebalancing bonus

Diversification quietly raises compounded return, not just lowers risk. A volatile path drags on growth: losing 10% then gaining 10% leaves you at 0.99, not 1.00. The long-run geometric return sits roughly a half-variance below the arithmetic mean:

g \approx \mu_p - \frac{\sigma_p^2}{2}

Here g is the geometric (compounded) return. Now compare a rebalanced blend against the average of holding each asset separately. The arithmetic mean return is the same either way (return is linear). But the blend's variance is smaller, so its half-variance drag is smaller. The difference is the rebalancing bonus:

\text{bonus} \approx \frac{\bar{\sigma^2} - \sigma_p^2}{2}

where \bar{\sigma^2} is the average of the assets' variances and \sigma_p^2 is the portfolio variance. This is exactly the variance you diversified away, cut in half.

Equal-weight bonus on the demo assets

  1. Average variance: (0.16^2 + 0.06^2 + 0.15^2) / 3 = (0.0256 + 0.0036 + 0.0225)/3 = 0.01723.
  2. Equal-weight (1/3 each) portfolio variance works out to about 0.00699 using the covariance formula above.
  3. Bonus: (0.01723 - 0.00699) / 2 = 0.00512, or about 0.51% per year of extra geometric return.

That half a percent is real, but only if you actually rebalance. Harvesting it means selling what rose and buying what fell on a schedule. Do nothing and the weights drift, so there is no bonus to collect.

With two assets at 7%/16% and 3%/6%, sliding the correlation from +1 down to -1 bends the frontier further left. At correlation +1 the frontier is a straight line between the two points; at -1 a specific mix reaches near-zero volatility.

Common mistakes

Trusting the exact weights. The optimizer answers your inputs to four decimals, but your inputs are guesses. Round the result and sanity-check it against nudged inputs.

Forgetting correlations move. The -0.1 stock-bond correlation that made diversification work has flipped positive in some stressed periods. When assets fall together, the frontier you drew no longer holds.

Assuming the bonus is free money. The bonus formula ignores transaction costs and taxes. Selling winners can trigger a tax bill. If rebalancing costs more than the half-variance you recover, skip it or widen the schedule. See the Capital Gains Tax Calculator for the tax side, and Investment Fee Impact Calculator for the cost side.

Confusing arithmetic and geometric return. The frontier plots arithmetic means. What you actually compound is the geometric return, lower by roughly \sigma_p^2 / 2. The Volatility Drag & Leveraged ETF Decay tool shows why that gap widens fast at high volatility.

Related tools

Once you have chosen an allocation, project it forward with the Investment Growth Calculator or a Dollar-Cost Averaging Calculator plan. Measure realized performance with the CAGR Calculator or the ROI Calculator, and strip out inflation with the Inflation-Adjusted Return Calculator. For cash-flow-based decisions, the NPV Calculator and IRR Calculator apply.

Frequently asked questions

Why does the tool sample 2000 random portfolios instead of solving exactly?

The random cloud makes the trade-off visible: you see how wasteful most mixes are. The corner portfolios (minimum variance and maximum Sharpe) are solved precisely; the cloud is there for intuition, not for finding the answer.

What correlation gives the biggest rebalancing bonus?

The lower the better. At correlation +1 the bonus is zero because no variance is diversified away. At -1 the portfolio variance can fall near zero, so the bonus is close to half the average variance. In the demo, dropping the stock-bond correlation from -0.1 to -0.5 would roughly double the equal-weight bonus.

How often should I rebalance?

Common schedules are yearly or when a weight drifts more than 5 points from target. There is no universal best. More frequent rebalancing captures the bonus faster but costs more in trades and taxes. Match the frequency to your costs.

Can the frontier tell me how much stock to hold?

It shows the efficient trade-offs, not your personal choice. Where you sit on the frontier depends on how much volatility you can tolerate, which no formula knows. This is an educational estimate, not financial advice; talk to a qualified adviser before committing money.

What happens with short selling allowed?

Dropping the long-only constraint lets weights go negative, so the frontier extends further in both directions. The math still works, but the risks change sharply: a levered short position can lose more than you invested. The tool assumes long-only weights.

Try it yourself in the tool

Related tools

CAGR Calculator

Compound annual growth rate between two values.

Dollar-Cost Averaging Calculator

Project a regular investing plan’s future value.

Investment Growth Calculator

Portfolio projection with monthly contributions.

Options Strategy Lab

Build multi-leg option strategies and see the payoff curve, Greeks and probability of profit update live.

ROI Calculator

Return on investment, simple and annualized.

Capital Gains Tax Calculator

Profit and tax on a sold investment.