Skip to content
Menu
Finabase Learn

Measure and manage portfolio risk Lesson 7 of 14

Portfolio and risk · Portfolio theory

Modern portfolio theory: why the mix matters more than the parts

Markowitz’s idea in plain terms: a portfolio’s risk depends on how its holdings move together. Every SPY and AGG mix from 2005 to 2026, measured.

On this page
  1. The idea
  2. How it works
  3. Using it
  4. Where it fails or misleads
  5. Compared with its neighbours
  6. How Finabase does it
  7. Frequently asked questions
  8. Summary
  9. Check your understanding
  10. Exercise

Modern portfolio theory is the idea that an investment should be judged by what it does to the whole portfolio, not by its own risk and return. Harry Markowitz set it out in 1952. The central observation is arithmetic: a portfolio's return is the weighted average of its holdings' returns, but its risk is less than the weighted average of their risks whenever the holdings do not move in step.

That gap between average risk and actual risk is the benefit of diversification, and it is the only benefit in investing that costs no expected return. This lesson shows it on two funds and then on three, and explains what the theory assumes, where it has held up and where it has not. It builds on the beginner lessons on Diversification: why spreading your investments reduces risk and Asset allocation: how stock, bond and cash mixes have behaved and puts numbers on them.

The idea

Suppose two investments each swing by 15% a year. If they always rise and fall together, a portfolio holding both also swings by 15%. If they move independently, some of one's bad months land in the other's good months and the portfolio swings less. Nothing about either investment changed. The risk reduction came entirely from combining them.

The figure plots every mix of a stock fund (SPY) and a bond fund (AGG) in steps of 10%, by the volatility and return each mix had over 2005 to 2026.

Every mix of SPY and AGG, 2005–2026: risk against return 0% 3% 6% 9% 12% 0% 3% 6% 9% 12% 15% 18% 100% bonds 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% stocks Volatility a year Return a year (CAGR) Percent labels: share in SPY, rebalanced monthly. Straight line: where the mixes would sit with no diversification.
Volatility and compound return of each mix of SPY and AGG, January 2005 to October 2026, monthly total returns, rebalanced monthly.

If diversification did nothing, the mixes would sit on the straight dashed line between the two ends. They sit to the left of it: every mix had less volatility than its weights alone would suggest. The effect is most visible at the bottom. Moving from 100% bonds to 10% stocks raised the return from 2.9% to 3.8% a year and volatility barely moved, from 4.5% to 4.6%.

How it works

The first line has no surprise in it: returns average. The second line is where the theory lives. The last term contains the correlation. With ρ = +1 the expression collapses to the plain weighted average of the two volatilities. With any lower value it is smaller. Because correlations between different kinds of assets are almost never +1, a mix is almost always less volatile than its parts would suggest. Correlation in a portfolio: what a correlation matrix tells you works through this formula term by term.

A worked example: every mix of SPY and AGG

The data: 262 monthly total returns of each fund, January 2005 to October 2026. SPY's volatility was 14.8% a year and AGG's 4.5%. Their correlation was 0.23.

Share in SPY Return a year (CAGR) Volatility Weighted average of the two volatilities Deepest fall
0% 2.9% 4.5% 4.5% −17.1%
10% 3.8% 4.6% 5.5% −15.8%
20% 4.6% 5.2% 6.5% −16.2%
40% 6.3% 7.0% 8.6% −21.0%
60% 7.8% 9.4% 10.7% −32.3%
80% 9.3% 12.1% 12.7% −42.2%
100% 10.7% 14.8% 14.8% −50.8%

Returns are after fund expenses, rebalanced to the stated weights every month; taxes and trading costs are not included.

Reading across any row, the actual volatility is below the weighted average. At 60/40 the weights alone say 10.7%; the portfolio delivered 9.4%. That 1.3-point difference is the diversification benefit for this pair over this period.

Two details are worth noticing.

  • The all-bond portfolio was not the safest. A portfolio with 10% in stocks fell less at its worst (−15.8%) than bonds alone (−17.1%). The lowest volatility of any mix was at about 2% stocks, a hair below the all-bond figure. Adding a little of the riskier asset lowered or held risk because the two did not fall at the same times.
  • The benefit shrinks towards the top. At 80% and 90% stocks the portfolio is dominated by one asset, and the actual and average volatilities nearly meet.

Three assets

Volatility of three assets and of an equal mix of them, 2005–2026 US stocks (SPY) 14.8% Long Treasuries (TLT) 13.6% Gold (GLD) 17.4% Average of the three 15.2% One-third in each (actual) 9.6% Annualized standard deviation of monthly returns; the mix is rebalanced monthly.
Volatility of SPY, TLT and GLD and of a portfolio one-third in each, rebalanced monthly, January 2005 to October 2026 (GLD: price only; it pays no income).

The effect is larger when the correlations are lower. Stocks, long Treasury bonds (TLT) and gold (GLD) each had a volatility between 13.6% and 17.4%. Their correlations with one another over the period were −0.09, 0.09 and 0.23, all close to zero.

Return a year Volatility Sharpe ratio Deepest fall
SPY 10.7% 14.8% 0.65 −50.8%
TLT 3.0% 13.6% 0.15 −47.3%
GLD 10.1% 17.4% 0.54 −42.9%
One-third each 8.6% 9.6% 0.73 −21.2%

The mix was less volatile than any of its three parts, its deepest fall was half that of the mildest of them, and its Sharpe ratio was higher than the best. Long Treasuries, taken alone, were the worst investment in the table: 3.0% a year with a 47.3% drawdown. Inside the mix they earned their place by falling at different times from the other two. This is the theory's most useful and least intuitive lesson: the question to ask of a holding is what it adds to the portfolio, and a mediocre asset with a low correlation can add more than a good one that moves with everything else.

Using it

Markowitz's full method turns these observations into a calculation. Given an expected return and volatility for each asset and a correlation for each pair, it finds, for any level of risk, the mix with the highest expected return. The set of those best mixes is the Efficient frontier: the best portfolios and why they are hard to find, covered in the next lesson.

Few individual investors run that calculation, and those who do find its output unstable. What carries over to practical decisions is the reasoning.

  1. Judge additions by correlation as well as by return. A new holding that behaves like the existing ones adds little, however good its record.
  2. Decide the mix first. The split between asset classes determines most of a portfolio's volatility and drawdown. In the table above, moving from 40% to 80% stocks doubled the deepest fall.
  3. Expect the parts to disappoint in turn. A diversified portfolio always contains something that is doing badly. In 2005 to 2026 that was foreign stocks and then bonds. That is the design working.
  4. Rebalance. The numbers above assume the weights are restored every month. Left alone, the faster-growing asset takes over and the portfolio drifts towards its risk. See Rebalancing a portfolio: calendar vs threshold, drift and taxes.

Where it fails or misleads

  • The inputs are forecasts. The formula is exact; the numbers fed into it are not. Expected returns are especially hard to estimate, and small changes in them produce large changes in the "optimal" mix.
  • Correlations change. Stocks and bonds had a correlation of 0.23 over the whole period, but it was not constant: in 2022 both fell together. A portfolio built on one regime's correlation meets another.
  • Volatility is the only risk it sees. The theory treats a month 5% above average as being as undesirable as a month 5% below, and assumes returns are roughly bell-shaped. Real returns have more extreme months than that.
  • Diversification weakens in crises. Correlations among risky assets rise in sell-offs. The 60/40 mix still fell 32.3% in 2007 to 2009.
  • One period is one sample. Every figure in this lesson is from 2005 to 2026, a period that was kind to US stocks and gold and unkind to foreign stocks. The pattern (the mix is less risky than the average of its parts) is general. The specific numbers are not.
  • It says nothing of costs, taxes or behaviour. Rebalancing has trading costs and may realize gains, and the best mix on paper is worthless if its holder sells at the bottom.

Compared with its neighbours

Idea What it adds Lesson
Diversification Do not hold only one thing Diversification: why spreading your investments reduces risk
Modern portfolio theory A formula for how much risk combining removes, given correlations This lesson
Efficient frontier The best available mix at each level of risk Efficient frontier: the best portfolios and why they are hard to find
CAPM What returns assets should offer if every investor follows the theory Capital asset pricing model (CAPM): what return a beta should earn
Risk parity Weighting by risk contributed instead of by money invested Risk parity: weighting a portfolio by risk instead of by money

How Finabase does it

Finabase has no portfolio optimizer and does not calculate a portfolio's volatility or the correlations among its holdings.

What it shows is the starting point for the reasoning. A portfolio page lists each position with its market value and draws an allocation bar of the largest holdings and cash, which shows at a glance whether one position dominates. The bar is by position, not by asset class or sector, so a portfolio of ten technology stocks looks as spread out as one holding ten different kinds of asset. The comparison view plots several symbols as percent changes from a common start; lines that separate and cross are holdings that diversify one another, and lines that move as one are not.

To reproduce the table in this lesson, export month-end prices for two funds from their History tabs, compute monthly percent changes, and apply the formula with the spreadsheet's STDEV.S and CORREL functions.

Frequently asked questions

What is modern portfolio theory in simple terms?

That the risk of a portfolio depends on how its holdings move in relation to each other, not only on how risky each is, and that combining holdings which do not move together reduces risk without reducing expected return.

Why is it called "modern"?

The name dates from the 1950s, when the approach replaced the practice of analysing each security in isolation. It is now more than seventy years old.

Does diversification lower returns?

It lowers the return compared with holding only the best-performing asset, which nobody can identify in advance. Compared with the weighted average of what the holdings earn, it costs nothing: returns average, while risk falls by more than the average.

Is the 60/40 portfolio "optimal"?

No single mix is. In the 2005 to 2026 data, mixes from 50% to 70% in stocks had the highest Sharpe ratios, with almost no difference among them. The right mix depends on how much loss its holder can tolerate, which the theory takes as given.

What are the main criticisms of the theory?

That it depends on forecasts no one can make accurately, treats volatility as the whole of risk, assumes stable correlations, and ignores extreme events. These are criticisms of using it as a calculator. The underlying insight about correlation is not in dispute.

Summary

Modern portfolio theory rests on one asymmetry: returns combine as a weighted average, and volatilities combine as something less, depending on correlation. Over 2005 to 2026 a 60/40 mix of SPY and AGG had a volatility of 9.4% where its weights implied 10.7%, and an equal mix of stocks, long Treasuries and gold had 9.6% where the average was 15.2%. An asset should be judged by its effect on the whole portfolio. The theory's weakness is its appetite for forecasts, which is why its reasoning has lasted better than its optimized answers.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. Two funds each have a volatility of 15% and a correlation of 0.2. What is the volatility of a 50/50 portfolio of them?

    1. Exactly 15%
    2. Below 15%
    3. Above 15%
    Show the answer

    B. Below 15% Portfolio volatility equals the weighted average only when the correlation is +1. At 0.2 it is well below: about 11.6% by the two-asset formula.

  2. From 2005 to 2026, long Treasuries alone returned 3.0% a year with a Sharpe ratio of 0.15. Why did adding them to stocks and gold raise the portfolio’s Sharpe ratio?

    1. Their return was the highest of the three
    2. Their correlation with the other two was near zero, so they reduced portfolio volatility by more than they reduced return
    3. Bonds cannot lose money
    Show the answer

    B. Their correlation with the other two was near zero, so they reduced portfolio volatility by more than they reduced return A holding is judged by its effect on the whole portfolio. An asset with a modest return and a low correlation can lower total risk enough to improve the ratio of return to risk.

  3. What is the weakest part of using modern portfolio theory as a calculator?

    1. The variance formula is approximate
    2. It needs forecasts of returns, volatilities and correlations, and small errors in them change the answer a great deal
    3. It works only for two assets
    Show the answer

    B. It needs forecasts of returns, volatilities and correlations, and small errors in them change the answer a great deal The mathematics is exact for any number of assets. The inputs are estimates, and optimized weights are very sensitive to them, particularly to expected returns.

  4. A portfolio moved from 100% bonds to 10% stocks and 90% bonds. In the 2005 to 2026 data, what happened to its volatility and return?

    1. Volatility rose sharply and return rose slightly
    2. Volatility was almost unchanged and return rose by about 0.9 points a year
    3. Both fell
    Show the answer

    B. Volatility was almost unchanged and return rose by about 0.9 points a year With a correlation of 0.23, a small stock position diversified the bonds. Volatility went from 4.5% to 4.6% while the return rose from 2.9% to 3.8% a year.

Exercise

One task to do in Finabase. Nothing is graded or saved.

Open the comparison of SPY, TLT and GLD and find the 1-year return row in the table. Average the three figures: that is the price return of a portfolio that started the year one-third in each, because returns combine as a weighted average. Which of the three did worst? Then look at the chart above the table: did the three lines fall at the same times?

Compare SPY, TLT and GLD
What you should find

The average always lands between the best and the worst of the three. Risk does not combine that way: in the lesson the three averaged 15.2% volatility and the mix had 9.6%, because their falls came at different times. Which fund is worst changes from year to year. One always is, and that is the mix working as designed. The figures are price only and one year is a small sample.

Sources and further reading

  1. Harry Markowitz, “Portfolio Selection,” Journal of Finance, vol. 7, no. 1 (1952)
  2. Harry Markowitz, Portfolio Selection: Efficient Diversification of Investments (Wiley, 1959)
  3. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1990 — press release (Markowitz, Miller, Sharpe)
  4. Finabase chart data — monthly bars and dividends of SPY, AGG and TLT, and monthly bars of GLD, January 2005 to October 2026; statistics computed by Finabase