Scribd Logo
Upload

Welcome to Scribd!

  • 27 views

    Single Exponential Smoothing Algorithm: 3.1 Notation

    Uploaded by

    RizkyIndra Permana
    This document discusses single exponential smoothing, an algorithm for forecasting time series data. It defines the notation used, describes how the algorithm works by updating forecasts based on errors from previous forecasts, and shows that forecasts are a weighted average of all past observations with more recent observations getting higher weights. It includes exercises for practicing applying the algorithm.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Single Exponential Smoothing Algorithm: 3.1 Notation

    Uploaded by

    RizkyIndra Permana
    27 views2 pages
    This document discusses single exponential smoothing, an algorithm for forecasting time series data. It defines the notation used, describes how the algorithm works by updating forecasts based on errors from previous forecasts, and shows that forecasts are a weighted average of all past observations with more recent observations getting higher weights. It includes exercises for practicing applying the algorithm.

    Original Description:

    just to know

    Original Title

    3SES

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses single exponential smoothing, an algorithm for forecasting time series data. It defines the notation used, describes how the algorithm works by updating forecasts based on errors from previous forecasts, and shows that forecasts are a weighted average of all past observations with more recent observations getting higher weights. It includes exercises for practicing applying the algorithm.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    27 views2 pages

    Single Exponential Smoothing Algorithm: 3.1 Notation

    Uploaded by

    RizkyIndra Permana
    This document discusses single exponential smoothing, an algorithm for forecasting time series data. It defines the notation used, describes how the algorithm works by updating forecasts based on errors from previous forecasts, and shows that forecasts are a weighted average of all past observations with more recent observations getting higher weights. It includes exercises for practicing applying the algorithm.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 2

    Chapter 3

    Single Exponential Smoothing Algorithm


    3.1 Notation
    From now on in the course we use the following notation:
    y
    1
    , y
    2
    , . . . , y
    n
    are the observed values of the time series
    y
    n
    is the last value of the series tobe observedi.e. we are currently at time n (months, quarters, years...)
    Forecasts for the value of the series at future times n+1, n+2, . . ., using a model tted to y
    1
    , . . . , y
    n
    , are
    denoted F
    n+1
    , F
    n+2
    , . . .. The k-step ahead forecast from time n would be F
    n+k
    .
    Fitted values using the model are F
    1
    , . . . , F
    n
    .
    The residuals or errors are y
    1
    F
    1
    , . . . , y
    n
    F
    n
    .
    3.1.1 Single Exponential Smoothing
    There is no obvious statistical model that we try to t (by regression or another tting technique). Exponen-
    tial Smoothing is simply an algorithm for creating forecasts iteratively on the basis of how well one did with
    previous forecasts.
    Suppose we make a forecast F
    t
    for the value of y
    t
    (which is not yet observed).
    Now we observe y
    t
    and wish to make a forecast F
    t +1
    . We do this by taking our old forecast F
    t
    and
    adjusting it using the error in forecasting y
    t
    as follows:
    F
    t +1
    =F
    t
    +(y
    t
    F
    t
    ),
    where is between 0 and 1.
    The nearer is to 1 then the larger the adjustment.
    We cannot forecast the rst term in the series (since F
    1
    =F
    0
    +(y
    0
    F
    0
    ) and there is no F
    0
    or y
    0
    ). By
    convention, we x F
    1
    = y
    1
    and only forecast y
    2
    onwards.
    16
    CHAPTER 3. SINGLE EXPONENTIAL SMOOTHING ALGORITHM
    '
    &
    $
    %
    Init: F
    1
    = y
    1
    and choose 0 <<1
    Forecast:

    F
    t +1
    =F
    t
    + (y
    t
    F
    t
    )
    Until no more observation are available then
    F
    n+k
    =F
    n+1
    , k 1
    Table 3.1: Simple Exponential Smoothing (SES) Algorithm.
    3.2 What does Exponential Smoothing Really Do?
    If we recursively apply the smoothing equation to F
    t +1
    , we get:
    F
    t +1
    = F
    t
    + (y
    t
    F
    t
    )
    = [F
    t 1
    + (y
    t 1
    F
    t 1
    )] + (y
    t
    [F
    t 1
    + (y
    t 1
    F
    t 1
    )])
    = y
    t
    + (1) y
    t 1
    +(1)
    2
    F
    t 1
    ,
    Now F
    t +1
    is in terms of y
    t
    , y
    t 1
    and F
    t 1
    . We can repeat this, replacing F
    t 1
    by F
    t 2
    +(y
    t 2
    F
    t 2
    ), to get
    F
    t +1
    is in terms of y
    t
    , y
    t 1
    , y
    t 2
    and F
    t 2
    . Doing this replacement another t 3 times, we end up with F
    t +1
    in terms of y
    1
    , . . . , y
    t
    and F
    1
    , and the following equation for F
    t +1
    :
    F
    t +1
    = y
    t
    + (1) y
    t 1
    + (1)
    2
    y
    t 2
    + + (1)
    t 1
    y
    1
    +(1)
    t
    F
    1
    (3.1)
    So exponential smoothing forecasts are a weighted sum of all the previous observations.
    3.3 Exercises
    (1) What is F
    t +1
    when =0? What happens as increases to 1? What range of values must F
    t +1
    lie in?
    (2) Here is a short time series. Calculate the exponentially smoothed series and make a forecast for the
    next value in the sequence, using =0.5 and =0.1:
    t y
    t
    F
    t
    (for =0.5) error F
    t
    (for =0.1) error
    1 3 3 0 3 0
    2 4
    3 2
    4
    (3) Can you make k-step ahead forecasts using exponential smoothing?
    (4) Which observation is given the biggest weight in the formula (3.1) for F
    t +1
    . Which is given the small-
    est? Is this sensible?
    17 Rozenn Dahyot 2011

    You might also like