Time Series Analysis Flashcards
Regression of a variable observed at different time periods.
time series analysis
limitation of time series analysis
- regression assumptions are unlikely to be met - e.g. residual errors are correlated
- the mean and variance may change over time
simplest time series
trend
simplest trend model
linear trend - dependent variable Y changes constantly with time (independent variable)
Why use a log-linear model?
If the residuals persist
what does a log linear model show?
exponential growth which has had a log applied to it, thus showcasing growth linearly
test for serially correlated errors
durbin watson test
time series model that regresses its own lags (past values)
autoregressive series
What is a covariant stationary series?
Time series where mean and variance remain the same throughtout
3 requirements of a covariant stationary time series
- expected value of the time series is constant and finite
- variance constant and finite
- covariance constant and finite
What is the standard error of autocorrelation?
1/sq rt(T) where t is the number of periods in data
series roughly stays within an average
mean reversion
What is the chain rule?
Use mean reversion to estimate the next period or 2 of values in a time series
whats the limitation of the chain rule?
It increases uncertainty because future estimates are used to calculate other values
two types of forecast errors?
in and out of sample forecast errors
why is out of sample forecast errors the preferred method?
allows precision of the forecasted values to be assessed - compares predicted values with actual values
guidelines for instability of regression coefficients
- dont mix data from 2 periods with different types of regulation
What is a random walk?
A time series with a predictable component and a random error component.
what time series smooths noise?
moving average
What is an autoregressive time series?
One that predicts future values based on past data.
What is a moving average time series?
Random noise fluctuations are smoothed over time to reveal a trend.
What is an ARMA model?
auto regressive moving average
What is the minimum data for ARMA?
80 observations
What is the ARCH model?
auto regressive conditional heteroskedasticity
5 possible scenarios with multiple time series
- neither time series has a unit root (use linear regression)
- reject null of unit root for independent variable (should not use linear regression)
- reject null of unit root for dependent variable (should not use linear regression)
- both series have unit roots but not cointegrated (should not use linear regression)
- both series have unit roots but are cointegrated (should not use linear regression)