Time Series -Introduction
Welcome to the next lesson of this module where we will cover the topic of time series analysis.
Within this, we discuss the different ways that time series data can be analysed through regression. Initially the OLS regression method is considered, which is the basis of simple regression models and instructs us of some of the essential assumptions that must be considered to produce valid regression models.
Additional complications to the simple regression are discussed, such as finite distribution lags. Alongside this, problems are noted, such as biases and the influence of collinearity.
The effects of autocorrelation are explained and considered and lead to the discussion of the ARIMA method which is commonly used for more complex forecasting procedures.
ARIMA assists in identifying additional patterns from within error terms left from other methods of time series regression. It makes use of autocorrelation of lagged errors as well as a moving average function. However it is dependant on some assumptions such as the stationarity of the data being used.
We explain how Augmented Dickey Fuller, ACF and PACF are used in order to determine the correct ARIMA model to be used. Finally there is consideration of methods of testing the success of the ARIMA model. This is followed by discussion of the case of panel data, which represents a more complex set of time series data.
Below are some goals and objectives for you to refer to after learning this section.
Goals for this section
- To understand the main regression methods for time series data.
To understand the assumptions and complexities relevant to applying these methods.
Objectives for this section
To be able to:
- Use and understand OLS.
- Have an appreciation of different matters affecting OLS, and the times when it can or cannot be applied.
- Understand the underlying assumptions of ARIMA
- Be able to determine the best ARIMA factors and apply the model
- Understand the methods of testing and assessing an ARIMA model
- Understand the complexity of panel data as a type of time series data set