Maximum Sharpe Ratio vs Minimum Semivariance Allocation
You have a candidate list, a date range, and a machine that will do the arithmetic. The only thing left is the decision that shapes everything downstream: which objective do you hand the optimizer?
In RSJ Portfolio that choice is a dropdown, and two of the entries sit at opposite philosophical ends of modern portfolio theory. Maximum Sharpe ratio asks: how much excess return do I get per unit of total risk? Minimum semivariance asks a narrower, more pessimistic question: how much do I lose on the days that actually hurt?
Both are available, both are selectable, and both will produce a defensible portfolio from the same candidate list — but not the same portfolio. This article is about picking between them. It is not about the statistics underneath semivariance; if you want that, the semivariance explainer covers the measurement in detail.
The Allocation Decision: Return-Seeking or Downside-Focused?
Every portfolio construction question collapses into a preference about risk. Not risk in general — which risk.
Maximum Sharpe ratio treats a +4% month and a −4% month as equally informative. It measures the spread of returns around the mean and rewards allocations that deliver more return for less spread. It is a symmetric objective, and it is the classic mean-variance answer to “what is the best portfolio I can hold?”
Minimum semivariance refuses the symmetry. It only looks at returns below a threshold — the losses — and minimizes the dispersion of those. A portfolio that occasionally spikes upward is not penalised for it. A portfolio that grinds downward on bad days is.
That difference in what gets counted produces genuinely different allocations from identical inputs. Sharpe will happily concentrate in a volatile asset with a strong expected return if the ratio works out. Semivariance will drift toward assets whose bad days are shallow and rare, even if their good days are unremarkable.
Your market outlook and your tolerance for drawdown decide this, not the math. The math is indifferent. So treat the objective as a statement of intent you make before you run anything.
The Landscape of Portfolio Objectives
RSJ Portfolio exposes five allocation methods, and the objective you pick is one of them, not the only one. From the methods and features page:
- Minimum volatility — minimise total portfolio variance, with no return input at all.
- Maximum Sharpe ratio — maximise excess return per unit of total risk.
- Minimum semivariance — minimise downside deviation only.
- Hierarchical risk parity — allocate by clustering the correlation structure rather than inverting a covariance matrix.
- Conditional value at risk — minimise expected loss in the tail beyond a threshold.
Maximum Sharpe ratio is the return-seeking end of that list. It is the only method in the set that explicitly needs a view on expected returns to do its job, which is why it is also the most sensitive to how good that view is.
Minimum semivariance sits at the downside-focused end. It is often usable without an explicit return forecast, because it never asks “how much will this earn?” — only “how badly does this behave when it falls?”
Layered on top of the objective are two more selectable inputs: the risk estimator (eight risk models are available) and the return model. Change either and the same objective yields a different allocation. That is not a bug; it is the reason the product exposes them. A Sharpe optimisation built on one risk model can disagree materially with a Sharpe optimisation built on another.
If you have not configured these yet, the first-run setup guide covers what is worth setting before your first run.
Selection Criteria: How to Choose Between Sharpe and Semivariance
Before comparing the two objectives on the numbers, compare them on the practical dimensions that decide whether you will actually keep using one.
Ease of use. Both are entries in the same objective selector in the same Windows application. There is no second tool, no separate install, no code. If you can pick minimum volatility, you can pick minimum semivariance.
Automation. RSJ Portfolio fetches end-of-day prices, runs the cleaning pipeline, and optimises in one workflow. Switching objectives does not mean re-preparing data — the candidate list and the cleaned price history stay put while you change the method. (For the data side in more depth, see automating end-of-day price updates.)
Security and data locality. Prices are held in a local cache and the optimisation runs on your machine. There is no requirement to upload your candidate list to a service to get an allocation back.
Price. Licensing is per user, not per method — the objective list is not tiered, so evaluating both costs you nothing beyond the run. Check the pricing page for exactly what the single user license includes.
Integration. Results are exportable, so the allocation can go into whatever you use for reporting, backtesting or further analysis. The input side is end-of-day data, which means this is a daily-cadence tool rather than an intraday one.
Verify your installer before the first run — the download page publishes a SHA-512 digest alongside a versioned filename:
cd ~/Downloads
sha512sum *.exe
# compare the printed digest with the SHA-512 published on
# https://portfolio.staging.rsj.de/downloads
Side-by-Side Comparison
| Maximum Sharpe ratio | Minimum semivariance | |
|---|---|---|
| Objective | Highest excess return per unit of total risk | Smallest dispersion of losses |
| Risk measure | Standard deviation — all deviations count | Downside deviation — only negative deviations count |
| Return assumption | Requires expected returns | Often usable without explicit return forecasts |
| Typical use case | You have a return view and want the best risk-adjusted portfolio | You care about drawdowns and want protection on the downside |
| Sensitivity to outliers | Symmetric: a large up-move and a large down-move both register as risk | Asymmetric: large up-moves are not penalised |
| Interpretation of results | “This is the most efficient portfolio for my return expectations” | “This is the portfolio with the mildest bad days” |
| In RSJ Portfolio | Selectable objective; combines with any of the eight risk models and the return model | Selectable objective; same risk models and constraints apply |
Two practical consequences fall out of that table.
First, Sharpe leans toward assets with high expected returns even when they are volatile, because volatility is the denominator and return is the numerator — a sufficiently large numerator wins. Semivariance leans toward assets with limited losses, because returns above the threshold simply do not enter the calculation.
Second, both objectives respond to the weight constraints you set. The constraints are part of the optimisation, not a post-processing step, so a cap on a single position changes the allocation for either method.
Verdict: Who Should Pick Which?
Pick maximum Sharpe ratio if you have return forecasts you trust and your goal is risk-adjusted performance in the conventional sense. It is the right objective when you are comparing candidate lists on efficiency rather than on comfort, and when you are willing to accept drawdowns in exchange for a higher expected reward per unit of variance.
Pick minimum semivariance if losses are the thing you are actually trying to control. If a bad quarter costs you your mandate, your sleep, or your conviction to stay invested, then the symmetric penalty of standard deviation is measuring the wrong thing for you.
Or do the obvious thing and run both. This is the part most comparisons skip. The two allocations are not competing answers to one question — they are answers to two different questions, and the gap between them is informative. Where they agree, you have a position that is robust to your choice of objective. Where they disagree, you have found the assets whose classification as “risky” depends entirely on whether you count the upside.
Because switching objectives in RSJ Portfolio is a dropdown change rather than a re-implementation, that comparison costs you two runs, not two days.
How RSJ Portfolio Optimizes Each Objective
The workflow is identical for both objectives, which is the point:
- Install and license. Run the Windows installer from the downloads page, verify the digest, and activate the license. Installation is covered in the guide.
- Configure once. Set data provider, local currency, date range, cache location and cleaning settings. The configuration chapter covers each. Multi-currency portfolios are converted along a defined path — see data and currencies.
- Build the candidate list. Search symbols and add them. The selection chapter walks through it.
- Choose the objective. Maximum Sharpe ratio or minimum semivariance.
- Pick the risk model and return model. Eight risk models are available; the choice changes the result for either objective.
- Set weight constraints.
- Run. Data retrieval, cleaning and optimisation happen in the workflow; results appear as an allocation, which you then export.
The hard way is writing the optimiser yourself — assembling the covariance or downside matrix, encoding the constraints, and debugging the solver before you ever see an allocation. The product’s way is one run per objective.
To keep the two runs comparable, record the parameters alongside the exports. Your own run sheet is enough; this is bookkeeping, not a product configuration file:
# run-sheet.yaml — your notes, kept next to the exported allocations
runs:
- label: sharpe-baseline
objective: maximum-sharpe-ratio
risk_model: "<the model you selected>"
return_model: "<the model you selected>"
constraints: "single-position cap, as configured"
date_range: "same for both runs"
- label: semivariance-baseline
objective: minimum-semivariance
risk_model: "<same as above>"
return_model: "<not used / same as above>"
constraints: "identical to the Sharpe run"
date_range: "same for both runs"
Then diff the two exports. A short script is enough — adjust the column names to whatever your export produces:
import pandas as pd
sharpe = pd.read_csv("alloc-max-sharpe.csv").set_index("symbol")["weight"]
semi = pd.read_csv("alloc-min-semivariance.csv").set_index("symbol")["weight"]
compare = pd.DataFrame({"sharpe": sharpe, "semivariance": semi}).fillna(0.0)
compare["delta"] = compare["semivariance"] - compare["sharpe"]
print(compare.reindex(compare["delta"].abs().sort_values(ascending=False).index).head(10))
print("positions only in one run:", compare[(compare[["sharpe", "semivariance"]] > 0).sum(axis=1) == 1].index.tolist())
The largest deltas are the names whose riskiness is objective-dependent. Those are worth a manual look before you commit capital.
FAQ
What is the main difference between maximum Sharpe ratio and minimum semivariance? Maximum Sharpe ratio seeks the highest excess return per unit of total volatility, using expected returns and standard deviation. Minimum semivariance minimises downside risk by focusing only on negative deviations, often without requiring return forecasts.
Can I use both objectives in RSJ Portfolio? Yes. RSJ Portfolio offers both maximum Sharpe ratio and minimum semivariance as selectable objectives, along with other methods such as minimum volatility and hierarchical risk parity.
Which objective is better for volatile markets? Minimum semivariance may be more suitable in volatile markets because it specifically targets downside risk, while maximum Sharpe ratio considers total volatility and may be more sensitive to return forecasts.
Do I need return forecasts for semivariance optimisation? Not necessarily. Minimum semivariance focuses on downside risk and can be implemented without explicit return forecasts, whereas maximum Sharpe ratio typically requires expected return inputs.
Conclusion
Maximum Sharpe ratio and minimum semivariance are not rivals with one correct winner. They encode two different definitions of risk, and the allocation you get back will reflect whichever definition you chose. If you have a return view you trust and want efficiency, take Sharpe. If losses are what you are managing against, take semivariance. If you are unsure, run both on the same candidate list and let the disagreement tell you where your portfolio is actually fragile.
RSJ Portfolio makes that a two-run exercise on a Windows desktop, with the data pipeline, risk models and constraints already in place. Start at portfolio.staging.rsj.de, or read the getting started guide for the parameters worth setting before your first optimisation.
Related posts
- First Run Setup: Portfolio Optimizer Configuration That Matters
- Portfolio Optimization Tools for Quants and Researchers
- Automating End-of-Day Price Updates for Portfolio Analysis
- Semivariance Explained: Measuring Only the Losses That Hurt
- How to Check That a Portfolio Optimizer Is Not Overfitting
- Minimum Variance vs Hierarchical Risk Parity: Which Fits?