# Tracking error

> Tracking error measures how much your portfolio's returns diverge from the benchmark's, day to day. It is the volatility of the gap rather than of the portfolio.

> 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).

Tracking error measures how far your portfolio wanders from the benchmark. Not
whether it did better or worse, but how much the gap between the two moves
around.

It is [volatility](/metrics/volatility) applied to the difference rather than
to the portfolio itself.

## How it is calculated

Gylder computes the **active return** for each day, which is your return minus
the benchmark's return on that day. Tracking error is the standard deviation of
those daily differences, annualised:

```
active return   = portfolio daily return − benchmark daily return
tracking error  = standard deviation of active returns × √252
```

A portfolio that mirrors the benchmark exactly has a tracking error of zero. An
index fund's would be a fraction of a percent. A concentrated portfolio of
individual holdings will be many percent.

## How to read it

Tracking error is directionless. It is equally large whether you diverged by
beating the benchmark or by trailing it, so a high figure tells you that you
are doing something different, not that it is working.

That makes it most useful paired with a measure of direction:

- With the [information ratio](/metrics/information-ratio), which divides your
  average active return by exactly this number. Tracking error is the
  denominator there, so the two are best read together.
- With [alpha](/metrics/alpha), for whether the divergence paid.

A near-zero tracking error against a benchmark you did not intend to track is
worth noticing, because it means your portfolio is doing roughly what the index
does. Whether that is what you want is your call, and Gylder does not have an
opinion on it.

## Caveats

The figure is only as meaningful as the benchmark. Measured against an index
that has little to do with what you hold, tracking error mostly reports that
the two are different things, which you already knew.

It also inherits the sampling limitation of every measure here. Deep in
history, your value series is stored weekly rather than daily, so a tracking
error computed over a long window is a smoothed approximation.