Time Series Data Flashcards
Time Series
Sequence of data points, typically measured at successive points in time.
True or false: Observations that are close together in time are generally more closely related than observations further apart.
True
Univariate time-series analysis
Using a single sequence of data for a given sequence of time periods
Multivariate time-series analysis
Using multiple sets of data captured for the same sequence of time periods
The components of a time series
- Period - Trend - Seasonal Variation - Cyclical Variation - Stationary - Noise
Period
The unit of the analysis which is equal to discrete time intervals when the data measurements are taken
Trend
The long-term movement of the time series. It could be increasing, decreasing, or stationary.
Seasonal Variation
When a time series depicts a repetitive pattern that is observed over some time horizon lag. These effects are typically associated with calendar or climatic changes.
Cyclical Variation
An upturn or downturn that is not tied to seasonal variation. These effects usually result from changes in economic conditions.
Stationary
When the data fluctuates about a mean.
Non-stationary
When the data does not fluctuate about a mean
Noise
Any non-repeating, non-specific pattern which exists for a short duration
What are the components that can be separated from a time series via decomposition?
Trend (T) Seasonal (S) Random (R) Sometimes Cycle (C)
What is the benefit of decomposing time series?
Decomposition allows for seasonal adjustment and more clear identification of trends.
Fill in the blank: The decomposed components of a time series are assumed to follow either a(n) ____ or a(n) ____ model.
Additive Multiplicative
When are additive model useful?
Additive models are useful when seasonal variation is relatively constant over time.
When are multiplicative models useful?
Multiplicative models are useful when seasonal variation increases over time.
Additive Model
yt = T + S + R
Multiplicative Model
yt = T * S * R
What do T, S, R, and C stand for in time series models?
T - Trend S - Seasonal R - Random C - Cycle
Forecasting
Process of estimating outcomes of events in a future state.
Fill in the blank: It is important to understand ____ and ____ while forecasting outcomes and one must indicate the degree of ____ attached to the forecasts.
Risk Uncertainty Uncertainty
Prediction Interval
An estimate of an interval in which a future observation will fall with a certain probability.
Which is narrower: The range of a 95% prediction interval or a 99% prediction interval.
95% prediction interval
What are the two types of forecasting approaches?
Qualitative Quantitative
Qualitative Forecasting Approaches
Subjective and based upon opinions and judgement of consumers or experts.
Quantitative Forecasting Approaches
Based upon analysis of data using mathematical techniques.
When are qualitative forecasting approaches more appropriate?
When past data are not available
When are quantitative forecasting approaches more appropriate?
When past data are available
Time Series Analysis for Forecasting
Using historical data as the basis of estimating future outcomes
What are five methods of time series analysis for forecasting?
Moving Average Weighted Moving Average Exponential Smoothing Autoregressive Moving Average (ARMA) Autoregressive Integrated Moving Average (ARIMA)
What are some questions you should ask when working with data?
What is the source of the data? How was the data recorded? When was the data recorded? Has the data been processed? Are there any missing data points? What units were used to measure the data?
Naive Model
The forecast value for the next observation is simply what was observed this time. Ft+1 = Yt
What are the four methods of measuring forecast errors?
BIAS - Arithmetic Mean of Error MAD - Mean Absolute Deviation MSE - Mean Squared Error MAPE - Mean Absolute Percentage Error
BIAS
Arithmetic Mean of Error ( Σ ( Error ) ) / n
MAD
Mean Absolute Deviation ( Σ | Error | ) / n
MSE
Mean Squared Error ( Σ ( Error^2) ) / n
MAPE
Mean Absolute Percentage Error 100 * ( Σ ( | Error | / Actual Value ) ) / n
Error
Actual Value - Forecast Value
Absolute Error
Error |
Squared Error
Error^2
Smoothing Time Series
Local averaging of data in order to smooth out any short-term fluctuations and highlight longer-term trends or cycles.
What is the most common technique of smoothing time series?
Moving Average
Moving Average
Replaces the underlying data series by either a simple or weighted average of a specified number of elements.
Simple Moving Average
An equally weighted mean of the previous k consecutive data points.
When would you use a smaller value of k for SMA?
When there are sudden shifts in the underlying data
When would you use a larger value of k for SMA?
When fluctuations are infrequent in the underlying data.
When should you use moving average?
Non-seasonal time series Short term prediction
Weighted Moving Average
A weighted mean of the previous k consecutive data points, in which recent observations are given more weight. Note: The sum of the weights should be 1
The first step in time series analysis is what?
Plot the data
