Time series analysis and forecast, Finding structure in short time series

Examples: Time series analysis

Finding structure in short time series

The ‘Caterpillar’-SSA technique allows to find structure in short time series. Natural decomposition of the time series is constructed on the base of the unique parameter (the window length). Grouping appropriate components of the decomposition, you can extract the trend, seasonality and its harmonic components. Then you can interpret found components severally, and put them together to construct interpreted approximation of the time series.

Figure 1.14: War: trend and annual periodicity

Figure 1.15: War: quarter periodicity and series approximation

Data: "War"
Description: U.S. combat deaths in the Indochina war, monthly, from 1966 to 1971
Source: M.F.Janowitz and B.Schweizer (1989) Ordinal and Percentile Clustering, Mathematical Social Sciences, V.18, P.135-186, Table 10.
Parameters: window length = 18
Figure 1.14: the time series components reconstructed from the eigentriples with numbers 1-2 (red line), the eigentriples with numbers 3-4 (blue line), and the initial time series
Figure 1.15: the time series components reconstructed from the eigentriples with numbers 1-8 (red line), from the eigentriples with numbers 5-6 (blue line), and the initial time series

Previous example | Next example