BSB123 Lecture 11: Topic 10 - Time Series Analysis and Forecasting
Time Series Analysis and Forecasting
- Time series data
oData collected through time on a regular basis
oEqual time intervals
oChronological order
oNo gaps
oConsists of four different components
Long-term trend (L)
Cyclical variation (C)
Seasonal variation (S)
Random variation (R)
oE.g.: GDP, CPI, exchange rate, stock prices, sales/profit, absentee rate etc.
- Objective of Time-series analysis
oDetect and quantify patterns from past data
oPatterns focus on trend and seasonal effects
oUse the patterns to forecast future values
- Long-Term Trend
oTrend – relatively smooth long-term increase or decrease of a series
oDuration is more than one year
oNot always linear
- Cyclical Effect
oCycle – wave-like pattern about a long-term trend that is generally apparent over a
number of years
oSeldom regular
oOften appear together with other components
oUp and downs around trend not related to consistent calendar events; duration > a
year
oVary in duration and amplitude
oNot necessarily repetitive
- Seasonal Effect
oRegular variation around trend based upon events consistent in each calendar year
Document Summary
Time series data: data collected through time on a regular basis, equal time intervals, chronological order, no gaps, consists of four different components. : gdp, cpi, exchange rate, stock prices, sales/profit, absentee rate etc. Objective of time-series analysis: detect and quantify patterns from past data, patterns focus on trend and seasonal effects, use the patterns to forecast future values. Long-term trend: trend relatively smooth long-term increase or decrease of a series, duration is more than one year, not always linear. Seasonal effect: regular variation around trend based upon events consistent in each calendar year, short, repetitive calendar periods, can be a day in a week or month in a year, duration less than one year. Random fluctuations (not focused on in this unit: irregular component based upon unusual or unpredictable events, tend to hide the existence of other, more predictable components, exists in almost all time series.
