Time series analysis - lecture 1 Time series analysis Analysis of data for which the temporal order of the observations is important Two major objectives: (i) Forecasting (ii) Retrospective analysis (trend and intervention analyses)
Time series analysis - lecture 1 Oil and coal consumption in the US and China
Time series analysis - lecture 1 Consumer price index by month in Sweden
Time series analysis - lecture 1 No. unemployed by month in Sweden
Time series analysis - lecture 1 No. air passengers by week in Sweden
Time series analysis - lecture 1 Nitrogen concentration in the Rhine River
Time series analysis - lecture 1 Time series decomposition – additive model Y t = T t + S t + e t, t = 1, …, n The trend component is a smooth function of time that may have a simple parametric form The seasonal component represents a pattern that is repeated every year (or period) The error terms are supposed to be independent and identically distributed with mean zero Trend Season Error
Time series analysis - lecture 1 Time series decomposition – multiplicative model Y t = T t * S t * e t, t = 1, …, n The trend component is a smooth function of time that may have a simple parametric form The seasonal component represents a pattern that is repeated every year (or period) The error terms are supposed to be independent and identically distributed with mean one Trend Season Error
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model MAPE Mean Absolute Percentage Error (MAPE) measures the accuracy of fitted time series values. It expresses accuracy as a percentage. where y t equals the actual value at time t, equals the fitted value, and n equals the number of observations.
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 Time series decomposition – multiplicative model
Time series analysis - lecture 1 No. unemployed men by month in Sweden
Time series analysis - lecture 1 Moving quarterly averages of monthly data
Time series analysis - lecture 1 Moving 1-year averages of monthly data
Time series analysis - lecture 1 Moving averages of purely random components
Time series analysis - lecture 1 Impact of moving averages Purely random errors are suppressed (variance reduction) Spurious trends may appear (serial correlations are introduced) The timing and abruptness of level shifts are influenced (abrupt changes are smoothed out and may be shifted in time)
Time series analysis - lecture 1 Single exponential forecast of the number of air passengers ( =0.2)
Time series analysis - lecture 1 Single exponential forecast of the number of air passengers ( =0.05)
Time series analysis - lecture 1 Scatter-chart of time-lagged observations - time lag one week
Time series analysis - lecture 1 Scatter-chart of time-lagged observations - time-lags one and two weeks
Time series analysis - lecture 1 Scatter-chart of time-lagged observations - time-lags 1, 2, 3, and 5 weeks
Time series analysis - lecture 1 Empirical autocorrelation function
Time series analysis - lecture 1 Empirical autocorrelation function
Time series analysis - lecture 1 Empirical autocorrelation function
Time series analysis - lecture 1 Information provided by empirical autocorrelation functions Noise level (high noise implies low autocorrelations) Long or short memory of past observations Long-term trends Seasonal patterns