Seasonality is when there are regular, repeated patterns that occur every year.

Instinctively, the most obvious seasonal patterns involve temperature variations (Summer is hot and sunny so cold drink sales increase) and holiday periods (Spending always increases dramatically around Christmas).

Visually (pictures here):

The intuition I want to give here is that seasonality involves patterns that are consistently above or below the mean of the series year after year. For that reason they can be spotted by eye. However the view may be obscured by a trend component, so the data will have to be Detrended before accurate analysis can begin.

There are different methods to model seasonality. They will be discussed in future posts, hit the Seasonality tag below to see all related articles.

The best starting point when beginning a forecast is to look at a graph of the data your working with. This serves to give you an impression of how the data behaves in a way that no other method can and will guide your procedure.

The first and most easy thing to spot in a series is the Trend:

(pictures coming)

I’m sure you know what a trend is but depending on your level of education, you may not know that trends do not have to be linear (straight lines) but can be quadtratic (curved), cubed (two changes in direction) and so on.  So for a beginner do not constrain yourself to think in terms of straight lines, even if that is all you have been taught.

For now I will just leave you with the visual impression, we will work with the maths once the other elements have been superficially explained and illustrated.

When we have any time series and want to make a forecast, we define certain characteristics about the series. These characteristics are by no means absolute and depending on your techniques, a different forecast will result.

As I have been taught, the four fundamental characteristics of any time series are:

Trend – The general tendency of the series.

Seasonality – Regular, repeated patterns that occur with a yearly time frame.

Cycle – Regular, repeated patterns that occur over a yearly time frame.

Outliers – ‘Freak’ results that are not assumed to be part of the normal process.

The culmination of these four characteristics is to remove any exploitable predictive structure in the data, and use it to make the most accurate forecast possible. Each characteristic will be explained in turn.

The method of forecasting I will be using is for Time Series data.  IE any process that changes over time.

I have been taught  two ways of doing this and there is an important distinction which I want to make clear at the start:

Forecasting:

Looking only at past values of the series, and through various assumptions which will be explained, projecting forward.

Econometrics:

Constructing a model of the time series based on the underlying processes and then projecting forward.

Forecasting therefore is kind of a dumbed down method from Econometrics, however it can be very sucessful in forecasting series where it may be too complicated (at this moment in time) to find all the component parts which motivate it.

Hello there!

My name is Mike and i’m a fresh Economics graduate about to become a  forecasting analyst for a major UK supermarket.

The aim of this blog is to record, explain and discuss forecasting techniques as I learn them.  Which will hopefully help anyone interested in entering the field as well as providing a place to get other peoples perspectives.

If you have any questions – feel free to ask!