Introduction
In this post I want to show one way to filter time series data using a custom Excel function. For the theory I rely heavily on a course by the Australian Bureau of Statistics which can be found here.So, why would you want to filter time series data? What is a filter? And what is a time series?
A time series is a collection of data points observed through time. For example, the monthly retail sales of clothing and footwear would be one such series.
A filter in this sense is a set of numbers that are applied to the time series to produce a moving average. So, for example, a filter could be the set of numbers (0.2, 0.6, 0.2). This would be applied to a data point and the points before and after (in the case of a symmetric filter) to produce an average i.e.:
y'(i) = 0.2*y(i-1) + 0.6*y(i) + 0.2*y(i+1)
The whole point of this is to remove frequencies or seasonality within the data. For instance with monthly time series you might want to remove events which occur every 12 months.
Mentioned in passing above was the fact that the example given is a symmetric filter. The other type is, unsurprisingly called an asymmetric filter. Symmetric filters have an odd number of weights and these weights are the same around the central weight i.e. w(i) = w(-i).
The macro
The easiest and most efficient way to apply this filter in Excel is by writing an array function in VBA. The code is shown below.
---------------------------------------------------------------------------------
Dim OutputRows As Long
Dim OutputCols As Long
Dim output() As Double
Dim vert As Boolean
Dim i As Integer
Dim j As Integer
Dim size As Integer
Dim width As Integer
Dim b As Double
Dim c() As Double
If TypeName(filter) = "Range" Then
If filter.Rows.Count > filter.Columns.Count Then
ReDim c(1 To filter.Rows.Count)
For i = 1 To UBound(c)
c(i) = filter(i, 1)
Next
Else
ReDim c(1 To filter.Columns.Count)
For i = 1 To UBound(c)
c(i) = filter(1, i)
Next
End If
Else
ReDim c(1 To UBound(filter))
For i = 1 To UBound(c)
c(i) = filter(i)
Next
End If
'Is the size of the filter odd?
width = Int(UBound(c) / 2)
If width = UBound(c) / 2 Then Exit Function
vert = False
size = dataRange.Columns.Count
If dataRange.Rows.Count > dataRange.Columns.Count Then
vert = True
size = dataRange.Rows.Count
End If
With Application.Caller
OutputRows = .Rows.Count
OutputCols = .Columns.Count
End With
ReDim output(1 To OutputRows, 1 To OutputCols)
'Check output is the same size as input
If dataRange.Rows.Count <> OutputCols And dataRange.Columns.Count <> OutputCols Then
Exit Function
End If
'Populate output with zeroes
For i = 1 To OutputRows
For j = 1 To OutputCols
output(i, j) = 0
Next
Next
'Filter data
For i = width + 1 To size - width
b = 0
'For j = i - width To i + width
For j = 1 To 2 * width + 1
If vert = True Then
b = b + dataRange(i + j - width - 1, 1) * c(j)
Else
b = b + dataRange(1, i + j - width - 1) * c(j)
End If
Next
If vert = True Then
output(i, 1) = b
Else
output(1, i) = b
End If
Next
SeasonFilter = output
End Function
---------------------------------------------------------------------------------
As you can see most of the macro is concerned with setting up the parameters. Only the part towards the end applies the filter. Creating array functions in VBA is covered in detail here.
An Example
For this example I use the clothing retail sales for the UK supplied by those lovely people at the ONS. The February 2013 data can be downloaded from here. The data covers the period from Jan 1988 to Feb 2013. Enough data to show what a simple filter can do.
In fact, the filter I'm using is a symmetric 13 term filter with all equal weights of 1/13. The weights add to one so that the average of the new series equals the average of the unadjusted data.
I'm assuming that you know how to use array functions. If you don't, read this article.
In fact, the filter I'm using is a symmetric 13 term filter with all equal weights of 1/13. The weights add to one so that the average of the new series equals the average of the unadjusted data.
I'm assuming that you know how to use array functions. If you don't, read this article.
Clothing retail sales index for the UK - ONS data |
As you can see the data has been smoothed with some cycles removed. This is just a simple example and more filters could be applied to remove different frequencies/cycles.
However, how do you know which cycles are being removed? The answer will be revealed next time when I provide an Excel macro to calculate the gain function.
Until next time!
However, how do you know which cycles are being removed? The answer will be revealed next time when I provide an Excel macro to calculate the gain function.
Until next time!