Moving Average Processes

A q-order moving average process, denoted MA(q) takes the form

MA(q) processThinking of the subscripts i as representing time, we see that the value of y at time i+1 is a linear function of past errors. We assume that the error terms are independently distributed with a normal distribution with mean zero and a constant variance σ2.

Topics

The mathematical proofs of some of the properties of Moving Average Processes are provided in Moving Average Proofs

References

Greene, W. H. (2002) Econometric analysis. 5th Ed. Prentice-Hall
https://www.scirp.org/(S(351jmbntvnsjt1aadkposzje))/reference/referencespapers.aspx?referenceid=1243286

Gujarati, D. & Porter, D. (2009) Basic econometrics. 5th Ed. McGraw Hill
http://www.uop.edu.pk/ocontents/gujarati_book.pdf

Hamilton, J. D. (1994) Time series analysis. Princeton University Press
https://press.princeton.edu/books/hardcover/9780691042893/time-series-analysis

Wooldridge, J. M. (2009) Introductory econometrics, a modern approach. 5th Ed. South-Western, Cegage Learning
https://cbpbu.ac.in/userfiles/file/2020/STUDY_MAT/ECO/2.pdf

9 thoughts on “Moving Average Processes”

  1. MA(q) is a linear function of past q terms of the errors.
    My confusion is that where are these errors coming from? I have not yet developed a model.
    To explain further:
    When starting to work on an ARIMA model, I have only the time series data. How do I get the errors to identify q for the MA component of ARIMA.

    Reply
    • Amar,
      1. MA(q) is a process. The instantiation of the process is a time series. Sort of like the relationship between a population that follows a distribution and a sample from this population.
      2. The errors don’t identify the q value. You can try to estimate q using other approaches (ACF and PACF), but in the end you may need to try several q values and see which one gives the best fit to the time series data that you have.
      3. For any specific value of q, the errors are the difference between the values predicted by the MA(q) process and the time series data. This is similar to the differences between the observed y values and the y values predicted by a linear regression.
      4. The added problem is what sort of errors are there for the first few elements in the time series since there are no time series terms prior to time = 1. Let’s look at the MA(1) process defined by y_i = ε_i – .4ε_i-1 with σ^2 = .25, as described in Example 1 of
      https://real-statistics.com/time-series-analysis/moving-average-processes/calculating-ma-coefficients-solver/
      What values should be used for ε_0? Once we determine what this value, finding the values for ε_i for i >= 1 is straightforward. As shown in Figure 1 of the above webpage, we arbitrarily set ε_0 = 0. This is not the only choice and other choices may have advantages. In any case, if you have a stationary process, then after some time, the initial choices for ε_0 (as well as ε_i for negative i) won’t have much of an impact of the fit.
      Charles

      Reply
  2. Hello Charles,

    According to the above equation, a MA is the linear function value of the past errors.
    This is different from the commonly-known formula, which is an average of the past values.
    Is there any connection between two or are they totally different from each other?

    I really appreciate any help you can provide.
    Thanks.

    Jaron Lee

    Reply

Leave a Comment