TIME-SERIES DATA (ANALYSIS/DESIGN)

Examples of Time-Series Data

Macroeconomists are the most common users of time-series data. Typically, they have monthly or quarterly observations on a variety of economic phenomena and are interested in relating, say, some policy instrument, such as the money supply, to some economic variable of interest, such as the growth of GDP. Similar analysis is undertaken by political economists and sociologists who are interested in the relationship of political variables, such as the political orientation of the coalition in power, to economic outcomes, or the relationship of economic variables, such as unemployment or inflation, to political outcomes, such as presidential popularity or the outcome of elections. Time-series data arise in many other areas, such as the study of conflict, where studies analyze the causes of defense expenditures over time. In all of these studies, analysts must be aware of the frequency of their data, be it daily, monthly, quarterly, or annually; different models are appropriate for data of higher or lower frequency.

This entry limits itself to discussing stationary time-series. Roughly speaking, a time-series is stationary if its various statistical properties do not change over time; in particular, the best long-run forecast of a stationary series is its overall mean. Nonstationary series involve much more difficult mathematics; causes of nonstationarity usually involve some type of time trend. Analysts must be careful that they limit their use of stationary methods to stationary time-series.

Although time-series data refer to observations on a single unit, they can be generalized to TIME-SERIES CROSS-SECTIONAL DATA, where analysts study time-series obtained on a variety of units, usually having annual observations on a set of countries. Time-series cross-section data have some time-series properties, and so many issues relevant in the estimation of time-series models are also relevant. PANEL data are also related to time-series data; there, the analyst usually has only a few repeated observations on a large number of individuals. Although panel data present estimation problems related to time-series issues, the data are sufficiently different that different models and approaches must be used.

...