Frequency

Time series usually have a frequency, such as daily, weekly, monthly or quarterly. We distinguish
between two kinds of frequencies: frequencies that are based on the common calendar, and frequencies
that are based on companies’ fiscal calendars, and which therefore vary depending on the company in
question. We refer to the former as fixed frequencies and the latter as fiscal frequencies.

When a signal accepts a frequency argument, it is for one of two reasons: In most cases, the signal
can produce time series with different frequencies, and you must specify which output frequency you
want. In a few cases, a signal transform accepts a frequency argument because it needs to know what
the frequency of the input time series is. Usually, it can be inferred from the time series itself,
but there are cases where that is impossible, and then the transform will use the supplied frequency
instead.

This section goes through the various frequencies supported in the Exabel DSL, and explains their
meaning.

Fixed frequencies

Fixed frequencies are frequencies which do not depend on the entity in question. For such
frequencies, we use Pandas’ frequency codes. Intraday frequencies are not supported. Here are the
most useful ones:

Fiscal frequencies

Fiscal frequencies are frequencies that are based on a company’s fiscal calendar, and they therefore differ between companies. These frequencies are based on the
fiscal periods companies use for accounting and reporting. When a time series has a fiscal
frequency, it usually means that there is one data point for every fiscal period, and that the date
of a data point is the last date in the fiscal period it represents. This may, however, be
complicated by missing data points or fiscal periods being represented by other dates than the last
date in the period (see alignment).

There are fundamentally three different fiscal frequencies: quarterly, semi-annual and annual. All
companies report annually, and in addition either quarterly or semi-annually. In the United States,
public companies are required to publish quarterly reports, but in many other countries companies
may choose to publish semi-annual reports instead. But even for companies that report
semi-annually, there are often time series available with quarterly frequencies from other sources.

Accordingly, all companies for which we have financial data, have an annual fiscal frequency, and in
addition quarterly or semi-annual frequencies, and often both.

Fiscal frequency arguments

Fiscal frequencies can naturally only be used in the context of a specific company. When a frequency
argument is supplied to a signal, the company in question in determined in one of the following
ways:

• If the signal is evaluated for a company, the fiscal calendar of that company is used.
• If the signal is evaluated for an entity (e.g. merchant, app) which is owned by a company, the
  fiscal calendar of that company is used.
• Some signals accept a fiscal_entity argument, and if it is supplied, that entity is used,
  regardless of the evaluation entity.

When a signal accepts a frequency argument, a fiscal frequency can be specified in any of the
following ways. Note that not all signals accept all the variants, and may not accept any fiscal
frequencies at all. This is specified in the documentation of the individual signal.