A COMPANION TO Theoretical Econometrics
Salient Features of US Macroeconomic Time Series Data
The methods discussed in this chapter will be illustrated by application to five monthly economic time series for the US macroeconomy: inflation, as measured by the annual percentage change in the consumer price index (CPI); output growth, as measured by the growth rate of the index of industrial production; the unemployment rate; a short-term interest rate, as measured by the rate on 90-day US Treasury bills; and total real manufacturing and trade inventories, in logarithms.1 Time series plots of these five series are presented as the heavy solid lines in Figures 27.1-27.5.
|
1960 1965 1970 1975 1980 1985 1990 1995 2000 Year Figure 27.1 US unemployment rate (heavy solid line), recursive AR(BIC)/unit root pretest forecast (light solid line), and neural network forecast (dotted line) |
|
Figure 27.2 Six-month US CPI inflation at an annual rate (heavy solid line), recursive AR(BIC)/unit root pretest forecast (light solid line), and neural network forecast (dotted line) |
|
Figure 27.3 90-day Treasury bill at an annual rate (heavy solid line), recursive AR(BIC)/ unit root pretest forecast (light solid line), and neural network forecast (dotted line) |
|
Year Figure 27.4 Six-month growth of US industrial production at an annual rate (heavy solid line), recursive AR(BIC)/unit root pretest forecast (light solid line), and neural network forecast (dotted line) |
 |
In addition to being of interest in their own right, these series reflect some of the main statistical features present in many macroeconomic time series from developed economies. The 90-day Treasury bill rate, unemployment, inflation, and inventories all exhibit high persistence in the form of smooth long-run trends. These trends are clearly nonlinear, however, and follow no evident deterministic form, rather, the long-run component of these series can be thought of as a highly persistent stochastic trend. There has been much debate over whether this persistence is well modeled as arising from an autoregressive unit root in these series, and the issue of whether to impose a unit root (to first difference these data) is an important forecasting decision discussed below.
Two other features are evident in these series. All five series exhibit comovements, especially over the two to four year horizons. The twin recessions of the early 1980s, the long expansions of the mid-1980s and the 1990s, and the recession in 1990 are reflected in each series (although the IP (industrial production) growth rate series might require some smoothing to see this). Such movements over the business cycle are typical for macroeconomic time series data; for further discussion of business cycle properties of economic time series data, see Stock and Watson (1999b). Finally, to varying degrees the series contain high frequency noise. This is most evident in inflation and IP growth. This high frequency noise arises from short-term, essentially random fluctuations in economic activity and from measurement error.
