Treynor and information ratio: advanced tools for comparing funds

Sharpe and Sortino answer “was the bumpiness worth it?” These two answer sharper questions: was the market risk worth it, and was the manager’s deviation from the benchmark worth it?

They are the two ratios professionals reach for when Sharpe stops discriminating.

Treynor: return per unit of market risk

Treynor = (return − risk-free rate) ÷ beta.

Structurally identical to Sharpe, with one substitution: the denominator is beta instead of total volatility.

That swap changes the question. Sharpe divides by all risk — including the fund’s own stock-specific wobble. Treynor divides only by market risk, on the reasoning that stock-specific risk is diversifiable and an investor holding several funds has already diversified most of it away.

When Treynor is the better tool: when the fund is one holding inside a diversified portfolio. The only risk it genuinely adds to your total is its market exposure, so return per unit of beta is the relevant efficiency measure.

When Sharpe is better: when the fund is the portfolio, or close to it. Then its idiosyncratic risk is your risk, and you want the measure that counts all of it.

⚠️ Treynor inherits every weakness of beta, which means it also inherits the R² caveat: if the benchmark explains little of the fund’s movement, beta is unreliable and Treynor is unreliable with it. Check R² first.

Information ratio: return per unit of deviation

IR = active return ÷ tracking error, where active return is the fund’s return minus its benchmark’s, and tracking error is the volatility of that difference.

Read it as: how much outperformance did the manager extract for each unit of risk they took by not being the index?

This is the purest measure of active management skill available, because it prices the deviation itself. A manager who beats the benchmark by 2% while veering wildly around it has done something less impressive than one who beats it by 1.5% steadily — the second is far more likely to repeat.

Rough interpretation, in ascending order: 0.5 is respectable, 0.75 is good, 1.0 and above is genuinely strong — and sustained IRs above 1 over long periods are rare. Negative means the manager destroyed value relative to just holding the index.

Which to use when

• Comparing several active funds in the same category against the same benchmark → information ratio. It directly measures who converted active risk into active return most efficiently.
• Assessing a fund as one sleeve of a diversified portfolio → Treynor.
• Assessing a fund that is most of your portfolio → Sharpe.
• Worried about downside specifically → Sortino, which only counts downside deviation.

All four are on every fund page here, and all four should be read within a peer group. None has a meaningful absolute value: a Treynor of 0.08 is excellent or poor depending entirely on what the category did.

Pitfalls to avoid

• Using Treynor on a fund with low R². Beta is not describing anything, so neither is Treynor.
• Reading IR against the wrong benchmark. A mid-cap fund measured against a large-cap index produces a flattering IR that is really a cap-size effect.
• Comparing IR across categories. It is benchmark-relative by construction.
• Trusting one year. These are noisy over short windows and mean-revert. Three to five years minimum.
• Chasing the highest IR. A very high IR over a short period usually reflects one concentrated bet that worked, not a repeatable process.
• Forgetting cost. These are computed on returns after the expense ratio, which is correct — but it means a high-cost fund needs more gross skill to show the same ratio, which is the active management problem in one sentence.

Key takeaway

Treynor prices market risk; the information ratio prices the decision to differ from the index. For picking between active funds in one category, IR is the most directly relevant number on the page — it isolates skill from both the market’s return and the fund’s willingness to take risk against its benchmark. Read both inside a peer group, over multi-year windows, and only after checking that R² makes the benchmark relationship meaningful in the first place.