RSJ Portfolio Create a statistically optimized portfolio

← All posts

2026-11-18

Minimum Variance vs Maximum Sharpe Portfolio: Which Is Better?

Minimum Variance vs Maximum Sharpe Portfolio: Which Is Better?
portfolio optimization modern portfolio theory portfolio correlation portfolio optimization hierachical risk parity minimize volatility minimize semivariance maximize sharpe

Minimum Variance vs Maximum Sharpe Portfolio: Which Is Better?

The comparison stops being academic the moment an optimizer asks you to pick an objective. There are two dropdown entries you will reach for first — minimum volatility and maximum Sharpe ratio — and they answer two different questions. One asks how little risk can this basket carry? The other asks how much return per unit of risk can I squeeze out of it?

Neither is “better” in the abstract. They consume different inputs, they behave differently when your estimates are wrong, and they produce portfolios that feel very different to hold through a drawdown. This article compares them on the things that actually drive the decision: required inputs, typical holdings, sensitivity to estimation error, concentration, and stability — and shows how RSJ Portfolio lets you select either objective and compare the results side by side.

What Each Objective Actually Optimizes

Both objectives are solved with modern portfolio theory: you give the optimizer a covariance matrix, a set of candidate weights, and a constraint set, and it searches for the weight vector that satisfies the objective.

Minimum variance: a pure risk model

Minimum variance minimizes portfolio volatility wᵀΣw using only the covariance matrix Σ of historic returns. There are no expected return inputs at all. The optimizer does not care whether a stock is expected to earn 8% or 20% — only how it moves relative to the rest of the basket.

That property is the whole appeal: the covariance matrix is estimated from data you already have (historic end-of-day prices), so the objective never consumes a forecast.

If you want the algebra — the Lagrangian, the closed-form weights, and why the inverse covariance matrix appears — the sibling article on minimum variance portfolio weights walks through it.

Maximum Sharpe: a risk-return model

Maximum Sharpe maximizes (wᵀμ − r_f) / √(wᵀΣw) — excess return over the risk-free rate, divided by volatility. It needs everything minimum variance needs, plus an estimate of expected returns μ for every candidate.

That extra input is the entire story. Here is the same trade-off in a few lines of Python, using the two objectives as they are usually written:

import numpy as np
from scipy.optimize import minimize

def min_variance(cov, bounds):
    n = cov.shape[0]
    x0 = np.repeat(1 / n, n)
    cons = ({"type": "eq", "fun": lambda w: w.sum() - 1},)
    return minimize(lambda w: w @ cov @ w, x0, bounds=bounds, constraints=cons).x

def max_sharpe(mu, cov, rf, bounds):
    n = cov.shape[0]
    x0 = np.repeat(1 / n, n)
    cons = ({"type": "eq", "fun": lambda w: w.sum() - 1},)
    return minimize(
        lambda w: -((w @ mu - rf) / np.sqrt(w @ cov @ w)),
        x0, bounds=bounds, constraints=cons
    ).x

Note the difference in signature: min_variance takes a covariance matrix and nothing else. max_sharpe takes mu. That one argument is where most of the practical difference between the two portfolios comes from.

Both objectives are also subject to weight constraints. Unconstrained, either one can put a large share of capital into a handful of names — the constraint set you configure is what keeps that in check.

Key Differences That Drive the Choice

Input sensitivity

Minimum variance needs a risk model. Maximum Sharpe needs a risk model and return forecasts. Return forecasts are notoriously noisy: small changes in μ can reorder the entire ranking of candidates, and the optimizer will happily exploit estimation error by loading up on whichever asset has the most optimistic number attached to it.

RSJ Portfolio ships eight risk models and a set of return models precisely because these estimates are a modelling choice, not a given. Testing the same candidate list under different risk estimators tells you how much of your allocation is signal and how much is a modelling artefact.

Behaviour in different markets

Because minimum variance ignores expected returns, it tends to settle on low-beta, defensive names and sectors whose price history shows small, steady moves. The portfolio often looks conservative — sometimes uncomfortably so in a strong bull market.

Maximum Sharpe is return-aware, so it tilts toward assets with attractive forecast returns relative to their risk contribution. In practice that can mean higher-beta names, and a portfolio that looks more growth-oriented than the minimum variance one.

Concentration

Both objectives are concentration-prone without constraints, for different reasons. Minimum variance concentrates in whatever is genuinely low-volatility and weakly correlated with the rest. Maximum Sharpe concentrates in whatever carries the best forecast return per unit of risk.

Weight constraints change both. Applied to minimum variance, they force the optimizer to spread out and accept slightly higher portfolio volatility in exchange. Applied to maximum Sharpe, they cap the optimistic names and often push the solution toward a broader basket — but the ranking underneath is still driven by forecasts.

Turnover and stability

Run the same optimizer on consecutive months and compare the weights:

  • Minimum variance weights typically move slowly, because volatilities and correlations are relatively persistent.
  • Maximum Sharpe weights can shift meaningfully whenever return estimates are refreshed, which means more rebalancing and more trading cost.

If you rebalance on a schedule, that difference matters. Automating rebalancing with an optimizer covers how to think about that cadence.

The practical implication

If you do not trust your return forecasts — or do not have a repeatable process for producing them — minimum variance is the safer objective. It removes an entire category of estimation error from the problem. If you have strong, disciplined views and want them expressed in the allocation, maximum Sharpe is the objective that lets them in.

How RSJ Portfolio Handles Both Objectives

RSJ Portfolio is a Windows desktop application that builds a statistically optimized portfolio from historic end-of-day prices, and the objective is a selectable parameter — not a separate product tier. The overview describes the overall workflow; in practice it runs like this:

  1. Build a candidate list. Search symbols and assemble the universe you want the optimizer to consider — see the selection chapter of the user guide.
  2. Choose the objective. Minimum volatility or maximum Sharpe ratio, alongside minimum semi variance, hierarchical risk parity, and conditional value at risk.
  3. Pick the risk model and return model. Eight risk estimators are available, so you can hold the candidate list and constraints fixed and change only the assumptions.
  4. Set the constraints and data settings. Weight constraints, date range, local currency, provider, cache, and the cleaning settings from the configuration chapter.
  5. Run the optimizer. The result is the recommended allocation, shown per position — documented in detail in the recommendation chapter.
  6. Export and compare. Because the objective is a parameter, you can run both on the same candidate list and diff the weights.

The data side is the part you would otherwise own yourself: provider connection, a price cleaning pipeline, a local cache, and the conversion path for portfolios quoted in several currencies. Solving the optimization directly, the way the Python snippet above does, is the hard way — you also inherit the data plumbing.

Installing the desktop app is a download and a SHA-512 check against the value published on the downloads page:

# compare against the SHA-512 published next to the versioned installer filename
Get-FileHash .\<installer>.exe -Algorithm SHA512 | Format-List

The methods and features page lists every objective, risk model, and constraint option in full.

Side-by-Side Comparison: Minimum Variance vs Maximum Sharpe

Minimum variance Maximum Sharpe
Objective Minimize portfolio volatility (wᵀΣw) Maximize excess return per unit of risk ((wᵀμ − r_f) / √(wᵀΣw))
Required inputs Covariance matrix only Covariance matrix plus expected returns, plus a risk-free rate
Typical holdings Low-volatility, low-beta, defensive names Assets with the best forecast return relative to risk contribution
Sensitivity to estimation error Moderate — driven by the risk model High — driven by return forecasts on top of the risk model
Diversification Correlated by design (weak correlation to the basket), but can concentrate Can concentrate in a few high-forecast names unless constrained
Behaviour under weight constraints Constraints spread the allocation and raise volatility slightly Constraints cap the optimistic names and broaden the basket
Stability of weights over time Usually more stable Can shift when return estimates are refreshed
Best suited to Risk-averse investors, capital preservation, no reliable return forecasts Return-focused investors with a disciplined forecasting process

A useful worksheet when you run both back to back — keep it next to the two exports:

comparison:
  candidate_list: core-europe-large-cap
  date_range: 2016-01-01/2026-09-30
  runs:
    - objective: minimum_volatility
      risk_model: <model-name>
      constraints: {max_position: 0.10}
      portfolio_volatility: <value>
      top_weights: [<symbol-a>, <symbol-b>, <symbol-c>]
    - objective: maximum_sharpe
      risk_model: <model-name>
      return_model: <model-name>
      constraints: {max_position: 0.10}
      portfolio_volatility: <value>
      top_weights: [<symbol-a>, <symbol-b>, <symbol-c>]
  decision: <minimum_volatility | maximum_sharpe | blend>

Fill the placeholders from the app’s result view; the point is to compare two runs that differ only in the objective.

Verdict: Who Should Pick Which?

Pick minimum variance if…

  • Capital preservation is the mandate, or drawdowns are the thing you actually lose sleep over.
  • You do not have a return-forecasting process you would defend in front of a committee.
  • You want weights that do not churn every time someone updates a spreadsheet.
  • You want the allocation to be explainable purely in terms of risk: “these names move independently and none of them is volatile on its own.”

Pick maximum Sharpe if…

  • You are managing for growth and accept that the allocation is a bet on your forecasts.
  • You have a disciplined, repeatable way to produce expected returns — and you re-run it consistently rather than eyeballing it.
  • You are comfortable with the possibility that the portfolio is concentrated in the names your model likes most.
  • You treat the output as one input to a decision rather than an answer.

Or run both

The most defensible workflow is not choosing once. Build the candidate list, fix the risk model and constraints, run minimum volatility, then run maximum Sharpe on the same inputs. Where the two allocations agree, you have a robust core. Where they diverge, you have a decision to make — and you now know exactly which assumption is driving it.

A hybrid is often the honest answer: weight the minimum variance solution more heavily when your confidence in return forecasts is low, and lean toward the maximum Sharpe solution when you have a genuine view you can articulate.

RSJ Portfolio is a single-user licensed Windows desktop application, which makes this kind of experimentation cheap: no per-run cost, no server to stand up, both objectives available on the same data with the same constraints. See pricing and license for what the license covers, and downloads for the installer and the user guide. Before trusting either output, it is worth reading how to verify portfolio optimization results — the checks apply equally to both objectives.

FAQ

Can I use both objectives in RSJ Portfolio?

Yes. RSJ Portfolio lets you select the optimization objective, including minimum volatility and maximum Sharpe ratio, so you can compare allocations directly on the same candidate list.

Does maximum Sharpe always outperform minimum variance?

No. Maximum Sharpe depends on expected return forecasts, which can be noisy. In practice, minimum variance can be more robust when return estimates are unreliable.

Which objective is more robust to estimation error?

Minimum variance is generally more robust because it only requires a risk model (covariance matrix) and does not rely on expected return estimates.

Can I apply weight constraints to both objectives?

Yes. RSJ Portfolio allows you to set weight constraints that apply to both minimum variance and maximum Sharpe optimizations, helping control concentration.

Which objective should I use for a first run?

Start with minimum volatility. It has one fewer input to get wrong, so if the result looks odd you know the problem is in the data or the risk model — not in a forecast. Once the data pipeline and constraints are validated, add the maximum Sharpe run and compare.

Conclusion

Minimum variance and maximum Sharpe are not competing answers to the same question. One minimizes risk with no view on returns; the other trades on a return view and pays for it with estimation error. If you cannot defend your forecasts, minimum variance is the more honest objective. If you can, maximum Sharpe puts that view to work.

The practical move is to stop arguing about it and run both. RSJ Portfolio exposes the objective as a parameter, ships eight risk models and a set of return models, and applies weight constraints to whichever objective you choose — so the comparison costs you one extra run, not a second toolchain.

Download the Windows installer, check the published SHA-512, and run the same candidate list through both objectives before you commit capital to either.

Related posts

EU label: AI-generated content