Time Serie - Seasonality (Cycle detection)

Regression with Season as Dummy Variable

The season is used a discreet regression variable and code it as dummy variables).

For instance:

• 1 if it's the season
• 0 if it's not the season

If you have more than one type of season, then you may need various dummy variables.

Seasonal Index

If you have five Decembers, then you should have five June. Having five of some month and six of other months will skew the seasonal indices.

Calculate a seasonal index for each month on past data.

If the average monthly sales is 100%, then the value in each month shows how that month compares to the average. Example:

• January can be 54.8% of the average month whereas December can 256.19% of the average month.

Do a normal regression and apply then this index to the forecast value.

Holt-Winters method

The Holt-Winters method was designed to handle data where there is a conventional seasonal cycle across the course of a year, such as monthly seasonality. However, many series have multiple cycles: the demand for electricity will have hourly (patterns across the hours of a day), daily (patterns across the days of the week), and monthly cycles. Similar patterns occur in the number of calls received by call centers or the workload faced by hospitals.

Fast Fourier Transform

FFT is a good tool to detect seasonality if we have a good amount of historical data. See http://nerds.airbnb.com/anomaly-detection

R Squared

A R-Squared value of 0.0999 will means that straight-line forecasting is not going to yield an accurate forecast.