How Does Portfolio Optimization Work?

By Augustine Hong on September 22, 2026

Key takeaways Optimization takes three inputs — expected return, volatility, and correlation — and solves for how much of your money goes into each holding.Correlation is the input most investors overlook, and the one that makes diversification work or fail.Mean-variance optimization is the core method; Black-Litterman, robust optimization, risk parity, and equal weighting each address a different weakness in it.The sequence is consistent: select the universe, gather inputs, run the model, apply constraints, review against your goals.No method predicts the future. What they do is narrow a messy set of possibilities to a defensible one.

Portfolio optimization sounds technical, and parts of it are. But the underlying idea is simple.

So how does portfolio optimization work in practice? It takes your assets, your goals, and a set of assumptions about risk and return, then calculates the mix that seeks the most from what you’re willing to risk. Everything else is detail.

Below is a plain look at the inputs, the methods, and the actual steps involved.

This article is educational and general in nature. It does not take into account the objectives, financial situation, or needs of any individual reader, and nothing in it is a recommendation to buy, sell, or hold any security or to adopt any investment strategy.

Figure 1 — How portfolio optimization actually runs: inputs, model, constraints, allocation. Illustrative only.
Source: Alexandria Capital, September 2026. Conceptual diagram of the optimization sequence. No data is plotted.

What Is Portfolio Optimization, Really?

At a basic level, portfolio optimization is the process of choosing how much money to put into each asset in your portfolio.

The goal is generally the same. Seek as much expected return as possible for a given level of risk, or as little risk as possible for a given return target. Every method described below is a different way of approaching that same equation.

Optimization is not a single formula. It’s closer to a family of related concepts, all built around the same tradeoff between risk and reward. Some approaches lean on decades-old math. Others use modern computing power to run thousands of scenarios in seconds.

None of these methods can predict the future. What they can do is narrow a large, messy set of possibilities to a smaller, defensible set of choices, even when the output looks like a single clean answer.

It also helps to know what optimization isn’t. It’s not a guarantee of higher returns, and it’s not a substitute for understanding your own goals and timeline. It’s a calculation layered on top of decisions you still have to make yourself.

The Inputs: Risk, Return, and Correlation

Before any optimization can run, it needs data. Three inputs matter most:

Expected return. What an asset is estimated to earn over time, based on historical performance and current conditions.
Risk (volatility). How much that return might swing, up or down, over the same period.
Correlation. How closely two assets move together. Low correlation is what makes diversification work.

These three inputs come together in something called a covariance matrix. It’s a table showing how every asset in your portfolio moves relative to every other asset. A large portfolio might have dozens of assets, which means the matrix has to account for hundreds of individual relationships at once.

This is where data analytics does the heavy lifting. Modern optimization tools pull years of price history, run the numbers, and surface a suggested allocation in seconds, work that would have taken a research team days to do by hand a few decades ago.

Correlation is the piece most new investors overlook. Two stocks in the same industry often move in the same direction for the same reasons. A utility company and a semiconductor maker tend to respond to different pressures entirely. A portfolio built without checking correlation can look diversified on paper, listing a dozen tickers, while behaving like a handful of bets on the same underlying trend.

Figure 2 — An illustrative correlation matrix across asset classes. Illustrative only.
Source: Alexandria Capital, September 2026. Hypothetical correlation estimates chosen to show the kinds of relationships that typically exist between asset classes. Not measured from market data.

FROM 30 YEARS IN THE CHAIR The most useful thing I can tell you about optimization is where it actually goes wrong, and it is almost never the arithmetic. It is the inputs. Every expected return in that model is somebody’s estimate, and estimates carry opinions. Early in my career I watched a well-built model produce a confident answer from assumptions nobody had examined in over a year. The model was fine. The inputs had gone stale. Now when I see a clean output, the first question I ask is not what the number says, it is who set these assumptions, when, and what would have to be true for them to still hold. — Augstine Hong

Mean-Variance Optimization: The Core Method

Mean-variance optimization is the original approach, and it’s still the foundation most other methods build on.

Economist Harry Markowitz introduced it in 1952.¹ The idea: use expected return, variance, and the covariance matrix to calculate an optimal allocation for every asset in the mix.

Mathematically, this is often solved as a quadratic programming problem. The model is seeking as much expected return as it can while holding risk within a set budget, or as little risk as it can while holding return above a set floor. Either way, it is solving for the same set of weights.

Plot every possible combination of assets by risk and return, and a curve appears called the efficient frontier. Portfolios sitting on that line are considered optimized. Portfolios sitting below it may be taking on more risk than necessary for the return they’re generating.

Picture a chart with risk along the bottom and expected return climbing up the side. Every dot represents one possible mix. The efficient frontier traces the outermost edge of that cloud. Where an investor should sit along that curve depends on their own tolerance for volatility, not on the math itself.

Mean-variance optimization does have a well-known flaw. Small changes in return assumptions can swing the resulting weights by a surprising amount.² Critics of modern portfolio theory point to this sensitivity, along with its assumption that markets behave rationally and that past volatility reliably predicts future risk. Correlations between assets can spike sharply during a downturn, right when diversification is needed most.

The Black-Litterman Model and Other Robust Optimization Methods

Later methods were built specifically to address mean-variance optimization’s weak spots.

The Black-Litterman model starts with market-implied expected returns, what the market as a whole seems to expect, and blends them with an investor’s own views.³ The result is a steadier set of weights that doesn’t overreact to a single aggressive forecast.

Robust optimization takes a different angle. Instead of assuming your inputs are exactly right, it builds in a margin for error, and looks for an allocation that holds up reasonably well across a range of possible outcomes rather than just the one you expect.

A few other approaches worth knowing:

Risk parity. Weights each asset by its share of total portfolio risk instead of by dollar amount.
Equal weighting. Splits money evenly across selected assets, skipping return forecasts entirely.
Monte Carlo simulation. Tests a proposed allocation against thousands of randomized market scenarios before it’s used with real money.

Table 1 — An illustrative comparison of portfolio optimization approaches.

Approach	How it works	Typically suits	Main limitation
Mean-variance optimization	Uses estimated returns, volatility, and correlations to calculate efficient asset weights	Portfolios with defined return targets and risk limits	Small changes in the inputs can produce large allocation changes
Black-Litterman model	Blends market-implied returns with selected investor views and confidence levels	Investors who want to incorporate forecasts without relying on them entirely	Results still depend on assumptions and confidence settings
Robust optimization	Tests allocations across a range of possible inputs instead of relying on one forecast	Situations where estimation error is a significant concern	Can produce more cautious allocations and requires greater computing complexity
Risk parity	Allocates according to each asset’s contribution to total portfolio risk	Portfolios seeking a more even distribution of risk across asset classes	May require frequent adjustment and, in some implementations, leverage
Equal weighting	Assigns the same percentage to every selected holding	Simpler portfolios with limited confidence in return forecasts	Ignores differences in volatility, expected return, and correlation
Monte Carlo simulation	Tests a proposed allocation across many randomized market scenarios	Evaluating how an allocation might behave across varied conditions	It is a testing method rather than an optimizer, and results depend on the assumptions used

Illustrative only. No approach is appropriate for every investor. Each method relies on assumptions and estimates that may prove inaccurate. Portfolio optimization does not predict or guarantee future results.

No single method of portfolio optimization suits everyone. The right one depends on portfolio size, how much complexity is worth managing, and how much confidence you place in the underlying forecasts.

How the Process Actually Runs

Strip away the math, and the process of portfolio optimization follows a fairly consistent sequence.

Selecting the universe. Determine which assets or financial instruments are eligible, such as a specific stock index, a bond universe, or a mix of both.
Gathering inputs. Pull historical return, risk, and correlation data for every asset under consideration.
Running the model. Apply mean-variance, Black-Litterman, or another method to calculate weights.
Applying constraints. Add real-world limits, like a position size cap or a minimum cash requirement, before finalizing the mix.
Reviewing and adjusting. Compare the suggested allocation against your goals and comfort with risk, then adjust before making investment decisions.

This is also where ongoing portfolio management comes in. Markets shift, so the inputs used to make decisions today won’t hold indefinitely. Most optimized portfolios get revisited on a set schedule rather than left untouched.

How often this sequence repeats depends on the portfolio. Some investors rerun it quarterly, others once a year, or only after a significant life event changes their goals. What stays constant is the sequence: gather data, run the model, apply real-world limits, and check the result against what you need.

What This Means for Your Portfolio

You may also come across a 70/20/10 rule while researching this topic. That’s a personal budgeting framework, 70% expenses, 20% savings, 10% debt or giving, not a portfolio allocation rule. It’s easy to mix the two up, but they answer different questions.

Portfolio optimization, by contrast, is specifically about how your invested assets are weighted against each other. It runs on the same core inputs, expected return, risk, and correlation, no matter which method calculates the final answer.

The math matters, but it isn’t the whole picture. A model is only as good as the person applying it and the assumptions it’s built on. That’s a good reason to understand the basic mechanics before relying on any single number.

Whether the calculation runs on a laptop, inside a robo-advisor, or on a portfolio manager’s desk, the same three inputs and the same tradeoff sit underneath it. Understanding that makes it easier to ask good questions about any recommendation you receive.

Frequently Asked Questions

What are the three inputs to portfolio optimization?

Expected return, volatility (risk), and correlation — how closely each pair of assets has moved together. Those three come together in a covariance matrix, which is what the model actually solves against.

Is portfolio optimization the same as diversification?

No. Diversification is spreading money across holdings that don’t all move together. Optimization is the calculation that decides how much goes into each one. Optimization is what turns diversification from a general principle into a specific set of weights.

Do I need software to optimize a portfolio?

Not to benefit from the thinking. A basic portfolio analysis — what you hold, how much risk each piece carries, and how it fits your stated goals — can reveal gaps a simple allocation chart misses. Formal modeling adds the most value as a portfolio grows in size and complexity.

If you would like to talk through how these ideas apply to your own circumstances, Alexandria Capital offers a complimentary consultation.

Notes and sources

1. Harry Markowitz, “Portfolio Selection,” The Journal of Finance, Vol. 7, No. 1 (March 1952), pp. 77–91. doi:10.1111/j.1540-6261.1952.tb01525.x

2. Richard O. Michaud, “The Markowitz Optimization Enigma: Is ‘Optimized’ Optimal?,” Financial Analysts Journal, Vol. 45, No. 1 (January/February 1989), pp. 31–42. doi:10.2469/faj.v45.n1.31

3. Fischer Black and Robert Litterman, “Global Portfolio Optimization,” Financial Analysts Journal, Vol. 48, No. 5 (September/October 1992), pp. 28–43. doi:10.2469/faj.v48.n5.28

About the author

Augustine Hong is Chief Executive Officer of Alexandria Capital, a Hightower company, and a Managing Partner at Hightower Advisors. Over more than 30 years he has advised high-net-worth families, family offices, founders, and institutions on capital allocation, governance, and risk, including in OCIO and OCFO capacities. He previously co-founded a $6 billion institutional structured investments platform and served as co-founder and chairman of the first mainstream Asian American cable network. He is a graduate of Binghamton University and completed the Advanced Management Program at The Wharton School.

Important disclosures

This material is provided by Alexandria Capital, a Hightower company, for educational and informational purposes only. It is not investment, legal, or tax advice, and it is not a recommendation to buy, sell, or hold any security or to adopt any investment strategy. It does not take into account the specific objectives, financial situation, or needs of any individual.

All investing involves risk, including the possible loss of principal. Diversification and portfolio optimization strategies do not ensure a profit or protect against loss in a declining market. Any figures, allocations, or charts shown are illustrative only and are not projections of, or targets for, any actual portfolio. Past performance is not indicative of future results.

Alexandria Capital is an investment adviser registered with the U.S. Securities and Exchange Commission. Registration does not imply a certain level of skill or training. Views expressed are those of the author as of the date shown and are subject to change without notice.