Timeseries Scaling

This guide explains how to rescale timeseries to a different calendar.

Introduction

Rescaling timeseries is the process of changing the changing the calendar of the TimeSeries and fitting the data using that new calendar.

Rescaling from a high frequency calendar to a low frequency calendar, e.g. Daily to Monthly is called aggregation

Rescaling from a low frequency calendar to a high frequency calendar, e.g. Monthly to Daily is called distribution

Aggregation

When you rescale data from a high frequency calendar to a low frequency calendar, we need to reduce or aggregate the data. The way the data is aggregated is determined by the observed setting.

Observed setting

beginning

An observation made at the beginning of the period

end

An observation made at the end of the period.

This is typically used for end of day settlement prices where the settlement value is either the last traded price or a calculation using the last tradded prices.

averaged

An average of the values observed throughout the period

summed

A sum of the values observed throughout the period

This is typically used for volumes, e.g. the number of traded items in a day, so aggregating to a lower frequency will sum up these values to give you a total volume for the period.

high

The highest value observed throuhout the period

low

The lowest value observed throughout the period

delta

The change of value from the start to the end of the period.

This is typically used for metered data where the values are meter readings which always increase in value. Scaling using delta allows you to see the amount used per period.

none

A point-in-time observation

Implicit observed

When you scale a TimeSeries without specifying the observed option, OpenDataDSL first checks the global observed setting which is set using:

set observed value

If this hasn't been set, it uses the observed attribute of the TimeSeries which defaults to end.

Distribution

When you rescale data from a low frequency calendar to a high frequency calendar, we need to distribute the observed value across a range of values. The way the data is distributed is determined by the distribution setting.

Distribution setting

constant

The value is constant across the whole period.

This can be seen like a bar chart where the tops of the bars are flat and the value jumps at each source period start.

linear

The values are interpolated using a linear spline between the source values.

This can be seen as a simple line chart where the values are points on the line

cubic

The values are interpolated using a cubic spline between the source values

This can be seen as a smoothed line chart.

none

A single value is used at the source index, the rest of the values are filled in with NaN or Missing

Implicit distribution

When you scale a TimeSeries without specifying the distribution method, OpenDataDSL first checks the global distribution setting which is set using:

set distribution value

If this hasn't been set then constant distribution is used.

Distribution value

The value used to calculate the distributed values is the observed value in the source TimeSeries except for the following cases:

Summed observed

If the TimeSeries is observed as summed then the distributed values are divided by the number of observations so that the sum of those observations matches the original observed value.

Scaling

To rescale a TimeSeries, you using the scale function which has 3 signatures:

ts = scale(input, calendar)
ts = scale(input, calendar, observed)
ts = scale(input, calendar, observed, distribution)

Examples

Monthly to Daily TimeSeries

// Create a monthly TimeSeries
mts = TimeSeries("2021-01-01", MonthlyCalendar(), [1,3,7,5,4,5])

// Scale to daily using a cubic spline
cts = scale(mts, DailyCalendar())
print cts.values

Monthly to Daily TimeSeries - cubic spline

// Create a monthly TimeSeries
mts = TimeSeries("2021-01-01", MonthlyCalendar(), [1,3,7,5,4,5])

// Scale to daily using a cubic spline
cts = scale(mts, DailyCalendar(), "beginning", "cubic")
print cts.values

Auto-scaling

By default, when the number of observations requested for a timeseries exceed a threshold, the timeseries is autoscaled according to the default observed setting on the timeseries.

The autoscaling thresholds depend on where the request comes from as shown in the table below:

Request Source	Observation Threshold
REST API and sdk's	25,000
Excel Add-in	25,000
ODSL in VS Code	25,000
Web Portal	5,000
Mobile Application	5,000

Detection

To determine if a timeseries has been autoscaled, there is a scaling property on the timeseries, if this is anything but the value 0 then it has been autoscaled.

Overriding

In order to tell the server not to autoscale a timeseries, you need to:

Request Source	Overriding
REST API	Add query parameter _scaling=0
Excel Add-in	Uncheck the option for autoscale
ODSL in VS Code	Use the command set autoscale off or add the option _scaling=0 to your data request
Web Portal	Check the scaling box and move the slider to the left