Time Series Final Exam Questions and Verified
Answers
What is a Time Series? Correct Answer: A collection of data points corresponding
to temporal measurements of some quantitative variable
Ex: Hourly website traffic, daily rainfall, monthly sales, quarterly revenue, annual
crime rates
T/F: Arbitrarily swapping rows in a time series will fundamentally change the data
Correct Answer: True
Define Time Series Analysis Correct Answer: Typically refers to modeling the
relationship between the y and time
What does a time series model characterize the relationship between? Correct
Answer: Between a point in time and all the points before it
Define forecasting Correct Answer: Make predictions using a time series model
With forecasting, the further into the future we go, the {more/less} certain we are.
This {widens/narrows} our prediction intervals. Correct Answer: less certain
widen our intervals
At a very general level, we can think of Time Series Analysis and Forecasting as:
Correct Answer: Trying to understand the past to predict the future
Univariate vs Multivariate Time Series Models Correct Answer: Univariate:
-future values of Y are forecasted using ONLY knowledge of past values of Y
Multivariate:
,- Future values of Y are forecasted using past values of Y AND one or more other
predict variables
When could adding a predictor variable be helpful? Correct Answer: A predictor
variable could be helpful if its pattern with time looks similar or inverse to the OG
relationship you're looking at.
AKA if it's correlated with the response variable.
What are the three important features of a time series? Correct Answer: 1. Serial
Correlation
2. Trend
3. Seasonality
Define Serial Correlation.
How is it quantified? Correct Answer: Serial correlation: the phenomena that
observations closer together in time tend to be more similar than observations
farther apart in time
Quantified by the autocorrelation function
- with data with serial correlation, we would see a high autocorrelation for a small
lag, and a small autocorrelation for a large lag
Autocorrelation of lag x Correct Answer: cor(y_t, y_{t+x})
How can we visualize autocorrelation / serial correlation? Correct Answer: We can
visualize the extent of autocorrelation in a given time series using ACF plots
With serial correlation, the height of the line on the ACF plot will generally
decrease as lag increases.
Define Trend
, What can it generally be approximated with? Correct Answer: Trend: the general,
smoothed behavior of a time series
- looking past subtle variations
- "squint your eyes, what is the time series generally doing?"
Can generally be approximated with low order polynomials
Define Seasonality
What is its connection to autocorrelation? Correct Answer: Seasonality:
characteristic of a time series in which the data experiences regular and predictable
fluctuations according to some period
If a time series experiences seasonality with period s, then observations s time units
apart are similar to one another
- "very strong autocorrelation across some period s"
Three parts of the Time Series Decomposition Correct Answer: 1. Trend
2. Seasonality
3. Random Variation
Effective time series models handle both _____ and ______ by accounting for
____________. Correct Answer: handle both trend and seasonality by accounting
for various autocorrelation structures in the observed data
T/F: With the right model design, random variation is avoidable Correct Answer:
False
Random Variation is unavoidable and is the chief contributor to "model
uncertainty"
Answers
What is a Time Series? Correct Answer: A collection of data points corresponding
to temporal measurements of some quantitative variable
Ex: Hourly website traffic, daily rainfall, monthly sales, quarterly revenue, annual
crime rates
T/F: Arbitrarily swapping rows in a time series will fundamentally change the data
Correct Answer: True
Define Time Series Analysis Correct Answer: Typically refers to modeling the
relationship between the y and time
What does a time series model characterize the relationship between? Correct
Answer: Between a point in time and all the points before it
Define forecasting Correct Answer: Make predictions using a time series model
With forecasting, the further into the future we go, the {more/less} certain we are.
This {widens/narrows} our prediction intervals. Correct Answer: less certain
widen our intervals
At a very general level, we can think of Time Series Analysis and Forecasting as:
Correct Answer: Trying to understand the past to predict the future
Univariate vs Multivariate Time Series Models Correct Answer: Univariate:
-future values of Y are forecasted using ONLY knowledge of past values of Y
Multivariate:
,- Future values of Y are forecasted using past values of Y AND one or more other
predict variables
When could adding a predictor variable be helpful? Correct Answer: A predictor
variable could be helpful if its pattern with time looks similar or inverse to the OG
relationship you're looking at.
AKA if it's correlated with the response variable.
What are the three important features of a time series? Correct Answer: 1. Serial
Correlation
2. Trend
3. Seasonality
Define Serial Correlation.
How is it quantified? Correct Answer: Serial correlation: the phenomena that
observations closer together in time tend to be more similar than observations
farther apart in time
Quantified by the autocorrelation function
- with data with serial correlation, we would see a high autocorrelation for a small
lag, and a small autocorrelation for a large lag
Autocorrelation of lag x Correct Answer: cor(y_t, y_{t+x})
How can we visualize autocorrelation / serial correlation? Correct Answer: We can
visualize the extent of autocorrelation in a given time series using ACF plots
With serial correlation, the height of the line on the ACF plot will generally
decrease as lag increases.
Define Trend
, What can it generally be approximated with? Correct Answer: Trend: the general,
smoothed behavior of a time series
- looking past subtle variations
- "squint your eyes, what is the time series generally doing?"
Can generally be approximated with low order polynomials
Define Seasonality
What is its connection to autocorrelation? Correct Answer: Seasonality:
characteristic of a time series in which the data experiences regular and predictable
fluctuations according to some period
If a time series experiences seasonality with period s, then observations s time units
apart are similar to one another
- "very strong autocorrelation across some period s"
Three parts of the Time Series Decomposition Correct Answer: 1. Trend
2. Seasonality
3. Random Variation
Effective time series models handle both _____ and ______ by accounting for
____________. Correct Answer: handle both trend and seasonality by accounting
for various autocorrelation structures in the observed data
T/F: With the right model design, random variation is avoidable Correct Answer:
False
Random Variation is unavoidable and is the chief contributor to "model
uncertainty"
