Year-Over-Year Approach

It is not uncommon to find some analysts who use a Year-Over-Year (YOY) approach to compare outcomes over two or more periods of time. This is sometimes accompanied by the use of Year-To-Date (YTD) data. This sometimes takes the form of accruing data from the beginning of one year to the present and comparing it to similarly accrued data from a prior year (or years).

There are probably more effective methods. Why?

Comparing YOY data, especially when based on YTD data, gives a signal filter whose bandwidth constantly changes as time progresses.

• At the beginning of the year, with just one month of data (beginning to end of January), the signal is sensitive to high frequency components (sometimes 'noise').

• Toward the end of the year have twelve months of data, and the high frequency components of the signal are attenuated (given much less weight).

Excerpts from Introduction to Time Series Analysis and Forecasting (Montgomery) section 4.1, page 173

• The smoother shown takes the average of the available data up to that point in time [crl: similar to the accrual of data by the YOY method]. For the constant process this smoother is quite effective [...]. What happens though if the process is not constant but exhibits a more complicated pattern.

• [Looking at a random walk pattern of Dow Jones returns over a longer time period]

• As the process changes, this smoother is having trouble keeping up with the process. The constant process assumption is no longer valid. However, as time goes on, the smoother accumulates more and more data points and gains some sort of "inertia". So when there is a change in the process, it becomes increasingly more difficult for this smoother to react to it.

• The constant process is not the norm but at best an exception.

• When there is a change in the process, earlier data no longer carry the information about the change in the process, yet they contribute to this inertia at an equal proportion compared to the more recent (and probably more useful) data.

• The most obvious choice is to somehow discount the older data.

An analogy:

• Taking a car trip. Listening to the radio over the duration of the trip. Constantly lower treble and increase bass.

• When start the trip, listen to the radio using all treble (no bass).

• When finish the trip, listen to the radio using all bass (no treble).

More effective methods might include:

• Regression (various approached)

• Time Series (various approaches)

• Statistical Process Control (SPC) (various approaches)