Maximum return means maximum risk. Optimization is about the best return for the risk you choose. This guide explains the efficient frontier, Sharpe and Sortino, correlations, exposure, crash liability and position sizing, with real data on what optimizers did and did not achieve.
If you only want maximum return, the answer is to hold the single riskiest asset and accept its worst falls. That is not optimization; it is a bet. Real optimization asks a better question: what is the highest return I can expect for the amount of risk I am willing to carry?
Harry Markowitz framed it this way in his 1952 paper “Portfolio Selection”: an efficient portfolio has the lowest variance for a given expected return, or the highest expected return for a given variance (Econlib biography of Markowitz). Plot every possible mix of a set of assets by risk and return, and the best mixes form a curve: the efficient frontier. Any portfolio below that curve could earn more for the same risk, or take less risk for the same return.
The raw materials show why no single asset wins on every measure:
| ETF (10 years to Sep 2026) | Return a year | Volatility | Sharpe | Sortino | Max drawdown | Correlation with SPY (60 months) |
|---|---|---|---|---|---|---|
| S&P 500 (SPY) | +15.2% | 15.3% | 0.86 | 1.37 | −23.9% | 1.00 |
| Nasdaq‑100 (QQQ) | +20.9% | 19.0% | 0.99 | 1.68 | −32.6% | 0.92 |
| Gold (GLD) | +11.7% | 15.5% | 0.66 | 1.16 | −23.8% | 0.17 |
| US bonds (AGG) | +0.9% | 5.2% | −0.21 | −0.28 | −17.1% | 0.62 |
| Long Treasuries (TLT) | −3.1% | 13.6% | −0.32 | −0.43 | −47.6% | 0.56 |
Source: Exent’s data, monthly total returns, as of Sep 30, 2026. Drawdowns are month‑end. The Nasdaq‑100 had the best return and the best Sharpe over this decade, and also the deepest fall of the stock funds. Gold earned less but moved almost independently of stocks, which is exactly what a frontier rewards. Past results do not predict future ones, and the next decade’s winners may differ.
The deeper theory is in mean‑variance optimization and the risk‑return trade‑off. One reason volatility matters even if you only care about growth is volatility drag: big swings lower your compound return, and a 50% loss needs a 100% gain just to get back to even.
You pick a level of risk you can live with, then improve the return per unit of that risk. The scores investors use for that are risk‑adjusted ratios. Each divides return by a different kind of risk:
| Ratio | Return divided by | What it rewards |
|---|---|---|
| Sharpe | Total volatility, after subtracting the cash rate | Steady returns in both directions |
| Sortino | Downside volatility only | Limiting losses while letting gains run |
| Calmar | The worst peak‑to‑trough fall | Surviving the deepest drawdown |
| Up/down capture | The market’s up and down months | Falling less than the market when it drops |
| Tracking error | How far you drift from a benchmark | Knowing how different your portfolio really is |
Beating the index on risk‑adjusted terms is hard. Of the 468 largest US stocks with five years of data, only 94 had a higher 5‑year Sharpe ratio than the S&P 500, in the same data as of Sep 30, 2026. Most single stocks carried more risk than they were paid for, which is the case for diversification in one number.
Go deeper on each part:
Partly. An optimizer finds the best mix for the past data it is given, and the future rarely repeats it exactly. A backtest ran two classic optimizers on six ETFs (SPY, QQQ, IWM, TLT, GLD and AGG). Each month they chose long‑only weights in 10% steps from the previous 36 months, then held them for a month. Two rule‑based mixes are shown for comparison.
| Backtest (hypothetical) | Period | Return a year | Volatility | Sharpe | Max drawdown | 2022 | S&P 500 Sharpe, same period |
|---|---|---|---|---|---|---|---|
| Max‑Sharpe optimizer | Jul 2019–Sep 2026 | +13.5% | 13.6% | 0.83 | −30.5% | −27.8% | 0.83 |
| Minimum‑variance optimizer | Jul 2019–Sep 2026 | +0.7% | 5.9% | −0.27 | −17.1% | −13.0% | 0.83 |
| Risk parity (inverse volatility) | Jul 2017–Sep 2026 | +8.7% | 10.4% | 0.65 | −23.0% | −18.2% | 0.82 |
| 60/40 stocks and bonds | Jul 2016–Sep 2026 | +9.5% | 10.1% | 0.75 | −20.0% | −15.8% | 0.88 |
Backtested results from ETF prices with dividends reinvested, monthly rebalancing and no trading costs or taxes. Hypothetical, with hindsight bias; not actual trading. Source: same data, as of Sep 30, 2026.
What optimization did and did not achieve here:
Academic research points the same way. DeMiguel, Garlappi and Uppal (2009) tested 14 optimization models on seven datasets and found none consistently beat a simple equal‑weight portfolio on Sharpe ratio (Review of Financial Studies). Estimation error in the inputs ate the theoretical gain. That is why practitioners add constraints, use longer data and treat the output as a guide.
More in why optimized portfolios often fail out of sample, risk parity and whether the 60/40 portfolio is dead. Every backtest above has its own page under strategies.
Assets that move differently from your stocks, especially in bad months. Correlation runs from −1 (opposite) to +1 (in step). The lower it is, the more a holding can cushion the rest. But correlations change when markets fall, so check them in stress months, not just on average.
| Asset | Correlation with S&P 500, all months | Correlation, calm months | Average return in 16 stress months | 2022 return |
|---|---|---|---|---|
| T‑bills (BIL) | 0.00 | 0.05 | +0.2% | +1.4% |
| Gold (GLD) | 0.14 | 0.10 | −0.4% | −0.8% |
| 7–10 year Treasuries (IEF) | 0.18 | 0.16 | −0.9% | −15.2% |
| Long Treasuries (TLT) | 0.19 | 0.18 | −1.9% | −31.2% |
| Commodities (DBC) | 0.34 | 0.28 | −1.2% | +19.3% |
| US bonds (AGG) | 0.40 | 0.32 | −1.2% | −13.0% |
| Minimum volatility stocks (USMV) | 0.87 | 0.77 | −4.9% | −9.4% |
| Nasdaq‑100 (QQQ) | 0.91 | 0.85 | −7.6% | −32.6% |
Monthly total returns from July 2016 to September 2026. Stress months are the 16 months when the S&P 500 fell more than 4%; it averaged −6.7% in those months. Source: same data, as of Sep 30, 2026.
Every asset’s correlation was higher over all months than over calm months alone: the bad months pulled things together. Cash and gold held up best in stress months. Treasuries helped in 2020 but fell with stocks in 2022. A low‑volatility stock fund fell less than the index but still moved with it.
The cluster pages cover each question:
Usually to fewer things than the number of holdings suggests. Exposure is the set of forces your money depends on: sectors, styles, the market’s direction, rates, currencies. Ten stocks in the same theme are one exposure, not ten.
Exent’s anonymized data on retail portfolios shows how common concentration is:
| Retail portfolios (anonymized) | Value |
|---|---|
| Portfolios in the sample | 144 |
| Median number of holdings | 15 |
| Median weight of the top five holdings | 77.2% |
| Portfolios with more than half the money in five holdings | 77.1% |
| Median weight in technology stocks | 22.6% |
| Portfolios with more than 40% in technology stocks | 28.5% |
| Portfolios holding at least one Magnificent 7 stock | 54.2% |
Source: anonymized sample, as of Sep 30, 2026. Internal accounts excluded; only aggregates of at least 20 portfolios are published.
Where to go next:
Replay your current holdings through real past crashes and turn the result into dollars. Volatility is an average; a crash is a single event, and it is what makes investors sell at the bottom. A portfolio that fell 35% in 2022 is a $105,000 paper loss on $300,000 (illustrative arithmetic), and the right question is whether you would have held through it.
Two measures help. A crash replay shows how each holding fell in a past crisis. Crash liability shows which holdings drove the portfolio’s losses in its worst months, which is often not the biggest position. A small, volatile holding that falls with everything else can do more damage than its weight suggests.
Size each position by the risk it adds, not by how much you like it. Investors commonly use one of four approaches, from simplest to most model‑dependent:
| Method | How it sets weights | Main weakness |
|---|---|---|
| Equal weight | Same amount in each holding | A volatile stock gets the same money as a calm one |
| Inverse volatility | Less money in holdings that swing more | Ignores how holdings move together |
| Risk budget | Each holding contributes a chosen share of total risk | Needs correlation estimates |
| Mean‑variance | Maximizes expected return for the risk chosen | Very sensitive to return estimates |
Whatever the method, many investors also cap single positions and rebalance when weights drift. None of this is a personal recommendation; the right weights depend on your goals and constraints.
Measure first, then change the few things that matter most. A practical order:
Hedging is the last step, not the first, because protection costs money every year it is not needed. The options, from diversifiers and cash to puts and collars, are in how to hedge a portfolio. To lower risk while keeping what you own, see how to reduce risk without selling.
It is choosing the mix of holdings that gives the highest expected return for the amount of risk you accept, or the lowest risk for the return you want. The set of best mixes is called the efficient frontier.
Above 1 is generally considered good and below 0 means the risk was not paid. For reference, the S&P 500 had a Sharpe ratio of 0.86 over the 10 years to Sep 30, 2026, in Exent’s data.
No. Higher expected returns come with more risk. What you can do is remove risk that is not paid for, such as several holdings that move as one bet, so you get more return per unit of risk.
Optimizers rely on estimates of future returns and correlations taken from the past. Small errors in those inputs lead to extreme weights, and the mix that was best in the past can do badly when conditions change, as stocks and bonds did in 2022.
Sometimes. In Exent’s data, long Treasuries rose 13.4% during the COVID crash of February and March 2020 but fell 31.2% in 2022, when stocks also fell. Their protection depends on why stocks are falling.
The count matters less than how the holdings move together. Fifteen stocks in one theme can behave like a few positions, so check correlations and sector weights, not just the number of holdings.
Exent measures your Sharpe, Sortino and Calmar ratios, correlation clusters and crash liability on your own holdings, in about a minute.
Educational content, not investment advice. Examples are illustrative; past performance does not guarantee future results.