Disclaimer: This is educational content about analytical software. Nothing here is investment advice, and trading carries a risk of loss. Paper trade any idea before you put money behind it.
Sharpe vs Treynor ratio: the short answer
Both ratios measure return earned per unit of risk, and both start from the same numerator: the average return you earned above the cash rate. The only difference is the denominator. Sharpe divides by total risk, the standard deviation of your returns. Treynor divides by market risk, your beta against a benchmark.
That one swap decides whether the ratio is usable. Treynor assumes the risk you didn't diversify away doesn't matter, because the rest of your wealth has already handled it. It also assumes beta is a stable, meaningful number. For a portfolio that is always invested, that's reasonable. For a strategy that holds a position on a minority of days, it usually isn't, and the rest of this post shows why with arithmetic you can check yourself.
If you only need the Sharpe side, our Sharpe ratio glossary entry covers the definition and the inputs. Here we focus on the comparison, and on the case other write-ups skip.
How we built the examples
A quick note on methodology, since it matters for how far to trust anything below. The tables and the 20-day series are arithmetic we worked out by hand and then re-checked in a script. They are not market data and not a record of any trading result. We built them so every intermediate number can be reproduced on a calculator, and we say so wherever a figure is illustrative. We find readers learn more from a series small enough to check than from a polished chart they can't reproduce.
The two formulas with the numerator spelled out
Write the excess return first, once, because it is shared:
Excess return = average strategy return per period minus the cash rate per period.
Then:
- Sharpe = excess return divided by the standard deviation of the strategy's returns.
- Treynor = excess return divided by the strategy's beta against a named benchmark.
Standard deviation counts every wobble, whether it came from the market or from your own signals. Beta counts only the part of your movement that tracks the benchmark. William Sharpe's own explanation of the ratio, hosted by Stanford at web.stanford.edu, frames it as a way to judge a fund against a benchmark, which is worth a read if you want the original reasoning rather than a summary.
For the cash rate, use the yield on a short-dated government bill, such as the Treasury bills listed on treasurydirect.gov. Pick one convention and keep it. We deliberately don't print a current rate here, because a number quoted today is stale next quarter.
How beta comes from a return series
Reports that cover beta estimation, including the longer explainers ranking for this topic, agree on the basics, and we found that the regression route is the same everywhere. This part is settled ground, so we'll keep it brisk. Beta is the covariance of your returns with the benchmark's returns, divided by the variance of the benchmark's returns. In a spreadsheet that's =SLOPE(strategy_returns, benchmark_returns). It's the same number you'd get from an ordinary least squares regression of your returns on the benchmark's.
Three practical points before we use it:
- Use excess returns on both sides if you can. With daily data and small rates the difference is minor, but be consistent.
- Match the dates. One benchmark return per strategy return, same calendar period, same frequency.
- Always report how many observations went into it. A beta from 5 data points and a beta from 500 are different animals, and the second half of this post is about exactly that.
One common misreading is worth flagging: someone estimates beta in a spreadsheet, gets a tiny number, and takes it as a sign of a market-independent edge. Per the definition, a small beta only says the returns did not co-move with that benchmark in that sample. It does not say the edge is real.
Why the division breaks near zero beta
Treynor puts beta in the denominator. Divide by a small number and the answer explodes. Several explainers state this in a sentence. Let's actually watch it happen.
Take one strategy with an average excess return of 8% a year. Nothing about its profit and loss changes in the table below. Only the beta we measure against changes:
| Measured beta | Excess return | Treynor ratio |
|---|---|---|
| 0.80 | 8% | 10% |
| 0.20 | 8% | 40% |
| 0.08 | 8% | 100% |
| 0.02 | 8% | 400% |
Same profit and loss, four answers, spanning a factor of 40. A beta estimate of 0.08 versus 0.02 is the kind of difference you get from adding or dropping a handful of days. So the ratio isn't telling you the strategy got better. It's telling you your beta estimate wobbled.
It gets worse on the other side of zero. A negative beta flips the sign, so a strategy that earned money and hedges the benchmark can show a negative Treynor, which reads as "bad" when it means the opposite. The ratio only behaves when beta is comfortably positive and stable.
Which strategies sit in the unstable zone by construction? Market-neutral and long-short books, where beta is hedged toward zero on purpose. Intraday strategies that close before the benchmark's overnight moves. Option sellers who run delta-hedged. And anything that spends most of its life flat, which brings us to the part we think matters most.
The out-of-the-market problem nobody mentions
Every page we read on this topic assumes a portfolio that is always invested. A systematic trader's strategy often isn't. If your system holds a position on 25% of days, then on the other 75% your return is zero, and zero doesn't move with the benchmark at all.
When you regress that return series on the benchmark across all days, the flat days drag the slope toward zero. The beta you get is largely the beta of cash. It describes what your equity curve did, not what your positions did while they were on.
Here is a small worked example. We built an illustrative 20-day series, not real trading results, with the benchmark's daily returns in percent:
0.4, 1.3, -0.4, -2.3, -1.0, -2.6, 0.6, 1.8, -1.1, -1.5, 1.9, 1.5, 0.7, -2.4, 0.3, 2.5, -1.6, -1.0, -2.3, 1.5
The strategy is in a position on only 5 of those 20 days (days 2, 8, 11, 16 and 20) and flat on the rest. On those five days its returns in percent are 1.2, -0.4, 2.6, 1.9 and -1.5. Every other day is 0. We'll use a cash rate of zero so the arithmetic stays visible.
- Average daily return across all 20 days: 0.19%.
- Standard deviation across all 20 days: 0.84%.
- Daily Sharpe: 0.19 divided by 0.84, about 0.23. Multiply by the square root of 252 and you get roughly 3.6 annualised.
- Beta across all 20 days: about 0.18.
- Treynor across all 20 days: 0.19 divided by 0.18, about 1.08 percentage points of daily return per unit of beta.
Now estimate beta only on the five position-holding days. The benchmark moved 1.3, 1.8, 1.9, 2.5 and 1.5 on those five days, and the strategy returned 1.2, -0.4, 2.6, 1.9 and -1.5. The slope comes out near 1.71.
So the same strategy has a beta of about 0.18 if you regress every day and about 1.71 if you regress only the days it held a position. That's nearly a tenfold difference, from the same trades. Treynor built on the first beta looks spectacular. Treynor built on the second is about 0.44 on a per-position-day basis, a different number, because it is answering a different question.
Two honest caveats. First, five points is far too few to trust any beta, and that's the lesson: report how many position-holding periods the beta came from, not just the beta. Second, the two Treynor figures aren't directly comparable, since one is per calendar day and the other per position day. What's comparable is the beta, and it moved by an order of magnitude.
The fix, in plain steps:
- Mark each bar as in-position or flat.
- Estimate beta using only the in-position bars.
- Publish the count of those bars next to the beta.
- If the count is small, don't quote a Treynor ratio at all.
Benchmark choice changes the answer
Beta is always beta against something. Regress the same strategy on a broad equity index, a sector index and a volatility index and you get three different betas, so three different Treynor values. As an illustration with made-up round numbers: with an 8% excess return, a beta of 0.15 against a broad index gives a Treynor of about 53%, a beta of 0.30 against a sector index gives about 27%, and a beta of -0.05 against a volatility index gives about -160%.
All three come from one strategy. A Treynor figure quoted without a named benchmark is not a figure. If a vendor, a forum post or a backtest report hands you one without saying which benchmark, ask or ignore it.
When Treynor is the right tool
We aren't saying Treynor is a bad metric. It's a good one for a specific job: comparing two well-diversified, mostly-invested long portfolios against one shared benchmark. There, the unsystematic risk really has been diversified away, beta is stable, and ranking by return per unit of market risk is the sensible question. CFA Institute's curriculum material on portfolio risk and return, available through cfainstitute.org, builds the same assumption in from the start.
Notice what that case isn't. It isn't a single-instrument system. It isn't a strategy that goes flat. It isn't a hedged book. The two ratios don't substitute for each other, because they answer different questions. Sharpe asks "how much did I earn per unit of everything that moved my equity curve?" Treynor asks "how much did I earn per unit of market exposure, assuming my other risk is irrelevant?"
Sampling frequency hits both ratios
Each ratio scales with the square root of the number of periods per year, so a daily figure and a monthly figure aren't comparable even for the same strategy. In our example, the daily Sharpe of 0.23 becomes about 3.6 when annualised with 252 trading days. Annualise the same underlying edge from monthly data with the square root of 12 and you'll get a different number, because the periods and their noise differ.
The rule is simple: only compare ratios that use the same frequency, the same window and the same cash-rate convention. We cover the annualisation arithmetic with the same care elsewhere on the site, starting from the Sharpe ratio entry, so we won't re-derive it here.
Which ratio answers which question
| Measure | Question it answers | What you must supply first |
|---|---|---|
| Sharpe | Return per unit of total volatility | Return series, cash-rate convention, sampling frequency |
| Treynor | Return per unit of market exposure | A named benchmark, and a beta estimated on position-holding periods with the count reported |
| Sortino | Return per unit of downside volatility | A minimum acceptable return, and enough losing periods to measure |
| Drawdown-based measures | Return relative to the worst peak-to-trough fall | The full equity curve in time order |
Notice that Treynor demands the most from you before it can be computed at all. That's the real reason most single-strategy traders should skip it: the preconditions aren't met.
What neither ratio can see
Three blind spots apply to both.
Path. A ratio compresses a whole equity curve into one number. One terrible week inside an otherwise smooth year barely moves it. Two strategies with the same Sharpe can have very different worst stretches, which is why we pair ratios with a look at the drawdown distribution instead of trusting one figure.
Fat tails. Standard deviation treats a loss that's 5 standard deviations away as if it were vanishingly rare. In markets it isn't. A strategy that sells insurance can show a lovely ratio right up to the day it doesn't.
The search itself. Run enough parameter combinations and the top ratio on the list is partly the search talking, not the strategy. Pick the winner of 500 backtests and you've picked the luckiest one as much as the strongest one. That's the territory of overfitting, and the practical defence is to test on data the search never saw, as described in the walk-forward testing entry. Sizing matters too: the ratio says nothing about how much you should risk per trade, which is a separate question covered under position sizing.
A practical checklist before you quote either ratio
- State the cash-rate convention (short-dated government bill) and the sampling frequency.
- For Sharpe, state the window and the number of observations.
- For Treynor, name the benchmark, estimate beta on position-holding periods only, and print the count.
- If the count of position-holding periods is small, drop Treynor and use Sharpe plus a drawdown view.
- If the number came from the top result of many tests, say so, and discount it.
None of this requires special software. It requires being honest about what the denominator really measures. Quantzee is analytical software, so it can calculate these inputs from your own results, but the judgment about which ratio fits your strategy stays with you.
Frequently Asked Questions
Source: William Sharpe's explanation of the ratio on web.stanford.edu. Our own illustrative series above is invented for teaching and is not a record of any live or backtested result.
