What are trends?
Trends are linear relationships over time, some trends are better seen in a growth rate (exponential growth) corresponding to a constant rate of growth, in which takes logs can be taken.
What is seasonality?
Seasonality captures regular “calander effects” on a time series. This can be seasonal, monthly, weekly, daily and hourly.
What are cycles?
Cycles refer to business cycles, not periodic functions, and are typically long duration patterns identified by looking at moving averages.
Describe how trends, seasonality and cycles are modelled.
Trends and cycles are often combined together as trend-cycles, Seasonality is always treated separately from these two.
What are step changes?
Step changes are abrupt changes in the level and or trend of a time series.
These typically correspond to some exogenous change in the system (policy change, major global event etc.)
What is down sampling how is it done?
Decreases resolution of data, can be random or not
What is up sampling how is it done?
Increases the resolution of the timeseries, uses interpolation
What are common data quality issues?
Completeness - missing data
Consistency - Outliers, sign errors
Overall these can be considered data reliability.
Accuracy and drift
Precision and measurement error.
What is currency>
The degree to which the data is current with the system it models
What is accessibility?
How easy it is to use the data you have
What are the fixes for data quality issues?
Interpolation - fill in blanks with reasonable numbers based on other data.
Padding - making methods work by filling the blanks with arbitrary values
What are the steps of a time series decomposition?
- Get the data
- Isolate the trend cycle component using moving average to smooth out the noise
2b. Calculate the detrended series by subtracting the moving average from the raw data leaving the detrended series.
- Estimate the seasonal component by averaging all of the detrended values for that time interval across all the years. Ensure they all sum to 0. The seasonal component is obtained by stringing together these monthly values and then repeating this sequence for each season.
- Calculate the residual component - subtract the seasonal series from the detrended series and what remains is the residual component (Should show no patterns and should be un correlated should also have mean 0 otherwise biased)
What does ARIMA stand for?
Auto Regressive Integrated Moving Average
