Function forecasting::WeightedMovingAverage(dataValues, estimates, noObservations, weights, noAveragingPeriods, ErrorMeasures, Residuals)

forecasting::WeightedMovingAverage

The weighted moving average procedure is a time series forecasting procedure. Essentially, this procedure forecasts by taking the weighted average over the last \(N\) observations.

Mathematical Formulation

Using the notation for observations and estimates given in Table 1, the estimates are defined as:

\[\begin{split}e_t = \sum_{j=1,\tau=t-(N+1)+j}^N {w_j \tilde y}_\tau \mspace{4mu}\mspace{4mu}\mspace{4mu} \textrm{ where } {\tilde y}_\tau = \left\{ \begin{array}{ll} y_1 & \textrm{ if } \tau < 1 \\ y_\tau & \textrm{ if } \tau \in \{1 .. T \} \\ e_\tau & \textrm{ if } \tau > T \end{array} \right.\end{split}\]

Function Prototype

To provide the error measures and residuals only when you need them, there are three flavors of the WeightedMovingAverage procedure provided:

forecasting::WeightedMovingAverage(
      ! Provides the estimates,
      ! but not the error measures nor the residuals
      dataValues,              ! Input, parameter indexed over time set
      estimates,               ! Output, parameter indexed over time set
      noObservations,          ! Scalar input, length history
      weights,                 ! Input, parameter
      noAveragingPeriods)      ! Scalar input, averaging length

forecasting::WeightedMovingAverageEM(
      ! Provides estimates and error measures, but not the residuals
      dataValues,              ! Input, parameter indexed over time set
      estimates,               ! Output, parameter indexed over time set
      noObservations,          ! Scalar input, length history
      weights,                 ! Input, parameter
      noAveragingPeriods,      ! Scalar input, averaging length
      ErrorMeasures)           ! Output, indexed over forecasting::ems

forecasting::WeightedMovingAverageEMR(
      ! Provides estimates, error measures, and residuals
      dataValues,              ! Input, parameter indexed over time set
      estimates,               ! Output, parameter indexed over time set
      noObservations,          ! Scalar input, length history
      weights,                 ! Input, parameter
      noAveragingPeriods,      ! Scalar input, averaging length
      ErrorMeasures,           ! Output, indexed over forecasting::ems
      Residuals)               ! Output, parameter indexed over time set

Here, the time set is a set that encompasses both the history and the horizon.

Arguments

dataValues

A one dimensional parameter containing the observations for the first \(T\) elements of the time set.

estimates

A one dimensional parameter containing the estimates for all elements in the time set.

noObservations

Specifies the number of elements that belong to the history of the time set. This parameter corresponds to \(T\) in the notation presented in Table 1.

weights

Specifies the weights. The weights should be indexed over a subset of Integers: \(\{ 1 .. N\}\), in the range \([0,1]\) and sum to 1.

noAveragingPeriods

Specifies the number of values used to compute a single average. This parameter corresponds to \(N\) in the mathematical notation above.

ErrorMeasures

The error measures as presented in Time Series Forecasting.

Residuals

The residuals as presented in Time Series Forecasting.

Example

With declarations and data as specified in Listing 1 the call:

weightSet := ElementRange(1,4);
locWeights := data { 1 : 0.1, 2 : 0.2, 3: 0.3, 4: 0.4 } ;
forecasting::WeightedMovingAverage(
    dataValues         :  p_dat,
    estimates          :  p_est,
    noObservations     :  91,
    weights            :  locWeights,
    noAveragingPeriods :  4);

This can be graphically displayed as:

Here the history is from 2021-04-01 till 2021-06-30 and the horizon is from 2021-07-01 till 2021-07-31.