# Rolling correlation (`ta.correl`)



<a id="ta-correl"></a>

Measure linear correlation between two price series on the same calendar.

[Indicator catalog](https://docs.openbacktesting.com/reference/indicators/index.md) · [How to use indicators](https://docs.openbacktesting.com/reference/indicator-api/index.md)

## Run this example

This complete script uses bundled data. It observes the selected output without placing trades.

```python
import numpy as np
from openbacktesting import Backtest, Strategy
from openbacktesting.test import GOOG

data = GOOG.copy()
data["benchmark"] = data.Close.shift(1)

class Example(Strategy):
    def init(self):
        self.value = self.ta.correl(inputs={'close': 'Close', 'benchmark': 'benchmark'})
        self.latest_value = None

    def next(self):
        channel = self.value
        if channel.valid[-1]:
            self.latest_value = float(channel[-1])

result = Backtest(data, Example).run()
print(result["_strategy"].latest_value)
```

Example inputs: benchmark is a lagged-price demonstration column; replace it with a real benchmark on the same calendar.

## Call and parameters

```python
self.ta.correl(*, length=30, inputs=None, backend=None, required_outputs=None)
```

Register in `init()`. Read the returned channel in `next()` after checking `valid[-1]`.

| Parameter | Meaning | Type | Default | Allowed values |
| --- | --- | --- | --- | --- |
| `length` | Calculation window in observations. | `int` | `30` | Minimum: `1` |

`source`/named source arguments accept a column name or same-run single-channel handle. Alternatively, `inputs` maps each input below to a column or handle. `backend` selects a supported implementation; `required_outputs` selects channels for readiness diagnostics. See [shared arguments and handles](https://docs.openbacktesting.com/reference/indicator-api/index.md#bind-inputs-and-read-outputs).

## Inputs and outputs

Required inputs: `close`, `benchmark`.

| Channel | Meaning |
| --- | --- |
| `correl` | Rolling correlation value |

## Reading values safely

Insufficient history or missing inputs can make a channel invalid. Check that channel's `valid` mask; `ready` is diagnostic and never suppresses `next()`. Multi-output results may have different valid channels on the same bar.

[Indicator catalog](https://docs.openbacktesting.com/reference/indicators/index.md) · [Custom calculations](https://docs.openbacktesting.com/reference/strategy/index.md#register-custom-calculations)