# Information ratio

> The information ratio divides how much you beat the benchmark by how erratically you did it. It measures consistency of outperformance, not size.

> These figures are derived from a return series that the app currently flags as being reworked.
> Values and holdings are unaffected. See [about these numbers](/metrics/about-these-numbers).

The information ratio asks whether your outperformance was consistent or lucky.

Beating a benchmark by 3% because you were steadily ahead all year is a
different achievement from beating it by 3% because one holding exploded in
March while everything else trailed. The information ratio separates the two by
dividing your average active return by how much that active return jumped
around.

## How it is calculated

```
information ratio = mean active return / standard deviation of active returns × √252
```

Both halves come from the same daily series: your return minus the benchmark's
return, day by day. The numerator is the average of those differences. The
denominator is their standard deviation, which is exactly
[tracking error](/metrics/tracking-error) before annualisation.

So the information ratio is your active return per unit of tracking error,
annualised with √252.

## How to read it

Positive means you were ahead of the benchmark on average across the window.
Negative means behind. The magnitude reflects how reliably, not how much.

A portfolio can have a large positive active return and a small information
ratio, which is the signature of outperformance concentrated in a few days.
That is worth knowing, because concentrated outperformance is much less likely
to repeat than steady outperformance.

Unlike the [Sharpe ratio](/metrics/sharpe-ratio), no risk-free rate enters
this calculation. The benchmark plays that role: what you are being measured
against is the alternative of simply holding the index.

Gylder does not define what a good information ratio is.

## What it misses

It treats active volatility symmetrically. Days you beat the benchmark by a
lot inflate the denominator exactly as much as days you trailed it by a lot, so
sharp outperformance is penalised alongside sharp underperformance.

It is also unstable over short windows, for the same reason
[alpha](/metrics/alpha) is: a small number of days and an estimated
relationship to a benchmark you chose. Over a few months this is close to
noise.