DSL: Switch to using Factset FX rates when doing currency conversions for Factset prices,
actuals, fundamentals, and estimates data. This adds support for more currencies (eg TWD)
than our previous data source (ECB, via Quandl). The converted values may change slightly
as the sources have slightly different ways of determining the closing FX rate for each date.
This only impacts signals that explicitly convert Factset data into a different currency.
DSL: Improved signal.change() (reference)
and signal.relative_change() (reference)
with smarter defaults to better handle irregular quarterly calendars.
Previously, quarterly time series with irregular quarter-end dates would have an unintuitive
result when calculating QoQ/YoY change, as the functions would forward fill the time series
to find the comparable data point. These changes now mean that the functions return the
intuitively expected results by default, but still give users options to override.
See our blog post for an example.
Technically, the changes are:
signal.change() and signal.relative_change() now take a method parameter which
controls how each data point is matched to a data point in the past. It defaults
to 'auto' which selects the method based on the offset.
The method parameter can be set to 'nearest' which matches each data point with the
one closest to its date minus the given offset. This is the default method when the
offset is a month or longer. When the offset is shorter, forward filling is the default.
Introduce a limit argument which controls how lenient we are when matching each
data point with the prior data point.
Note that the legacy relative_change(signal, ...) function was not changed, for
backwards-compatibility of existing signals that use it.
DSL: Deprecate the relative_change(signal, ...) function, which is now superseded by
signal.relative_change(). A deprecation warning will be returned when using this function.
As this function is widely used, we will only look to remove this function in September 2023.
As part of this change, we also removed the use_ffill_for_missing argument, which was not
used in any user signals.
Fix an edge case where relative_change(), signal.change() and signal.relative_change()
did not handle NaN values in the input data correctly when evaluated for multiple entities.
This could lead to incorrect results if the different entities had data with different
indices.