MODELS WITH TREND
Learning Objectives
1. Formalize simple models of variables with a time-dependent mean
2. Compaze models with deterministic versus stochastic trends
3. Show that the so-alled unt root problem arises in standard regression and
in times-series modes.
4. Explain how Monte Carlo and simulation techniques ean be used to derive
ciilical values for hypothesis testing
5, Develop and illustrate the Dickey-Fuller and augmented Dickey—Puller
tests forthe presence of a unit root
6. Apply the Dickey-Fuller tests to U.S, GDP and real exchange rates
7. Show how to apply the Dickey~Fuller testo series wih serial correlation,
roving average terms, multiple unit roots and seasonal unit roots
8, Consider tests for unit roots i the presence of structural change.
9, Illustrate the lack of power of the standard DickeyFullr test,
10. Show that generalized least squares (GLS) detrending methods can enhance
the power of the Dickey-Fuller tests.
AL. Explain how to use panel unit root tests in order to enhance the power ofthe
Dickey-Fuller test
12, Decompose a series with a tend into its stationary and tend components
1, DETERMINISTIC AND STOCHASTIC TRENDS
is helpful to represent the general solution to @ linear stochastic difference equation,
as consisting of these three distinct parts
y; = tend + stationary component + noise
Chapter 2 explained how to model the stationary component using the Box—
Jenkins methodology. Chapter 3 showed you how to model the variance of the error
(c., noise) component. A critical task for applied econometicians is to develop
simple stochastic difference equation models that mimic the behavior of wending
variables, The file RGDPXLS contains the quarterly values of real U.S, GDP over
the 1947Q1-201204 period (in billions of year 2005 dollars), From the plot of the
data shown in Figure 4.1, itis clear that the distinguishing feature real GDP {rgdp,)
181182 cHarTeR4 MODELS WITH TREND
17500:
15000.
12500
1000:
7500:
5000)
2500.
950 1960-1970 1980—«*1990.~—-2000--2010
— Actual Fited
Forecasts
FIGURE 4.1. A Deterministic Trend in Real GDP
is that it increases over time. For such a series, a naive forecaster might estimate the
sustained increase using the following cubic polynomial model for the trend:
redp, = 1890.247 + 9.1081 + 0.170 — 0.00017)
(27.66) 4.09) (870) (2.07) ap
‘The fitted values are shown asthe dashed lines in the figure, and the forecasted val-
tues are shown as the solid line extending past 201204. Regardless ofthe r-statistics, the
use of such a model for the trend of real GDP is problematic. Since there are no stochas-
tic components in the trend, (4.1) implies that there is a deterministic long-run growth
rate of the real economy. The “Real Business Cycle” school argues that technological
advancements have permanent effects on the trend of the macroeconomy. Since techno-
logical innovations are stochastic, the wend should reflect this underlying randomness.
Aaherents to other schools of macroeconomics would also argue that the trend is not
completely deterministic. For example, they might point out that an oil price shock or
a targeted tax reduction could affect investment and the economy's long-term growth
tale, Moreover, the implications for the behavior ofthe business cycle are not credible,
‘The deterministic trend implies that, whenever real GDP is below trend, in subsequent
periods, there will be unusually high growth as real GDP returns tothe trend. The reac
tion tothe 2067-2008 financial crisis suggests that most economists and politicians do
not take this notion very seriously. In fact, the forecasts beyond 2012 seem to totally
ignore decline in GDP resulting from the financial crisis. Lastly, the negative coefficient
‘on the P term implies permanent declines in future GDP after a sufficiently long time.DETERMINISTIC AND STOCHASTIC TRENDS 183
Reexamine the 7-month T-bill rate and the yield on 5-year U.S. federal govern-
‘ment securities shown in Figure 3.4. The (wo interest rates have no obvious tendency
to increase or decrease. Moreover, there are no decided structural breaks that induce
one-time shifts in the mean. Nevertheless, there is no pronounced tendeney for either
series to revert to a Iong-run mean, The Key feature of a trend is that it has a permanent
effect on a series. If the trend is defined as the permanent or nondecaying component
ofa time series, the two interest rates have a trend.
Suppose that a series always changes by the same fixed amount from one period
(o the next, To be more specific, suppose that
Ay = ay
As you know from Chapter 1, the solution to this linear difference equation is,
Y= Yo + aol
where yy is the initial condition for period zero,
Hence, the solution for Ay, = ay turns out (o be nothing more than a deterministic
linear time trend; the intercept is vq and the slope is ap. Now, if we add the stationary
component A(L}¢, to the trend, we obtain
Y= Jot aot + ALE, an
In (4.2). , can differ from its trend value by the amount A(L)¢,. Since this deviation
is stationary, the (),} sequence will exhibit only temporary departures from the trend.
As such, the long-term forecast of y,,, will converge tothe trend line yp + ao(t + 8). In
the jargon ofthe profession, this type of model is called a trend stationary (TS) model.
‘Now suppose that the expected change in y, is dp unis. In particular, let Ay, be
equal toa plus a white-noise term:
Ay, = ao te 63
Sometimes, Ay, exceeds dp and sometimes it falls short of ap. Since E,_¢, = 0.
(4.3) implies that y is expected to change by ay units from one period tothe next. The
seemingly innocuous modification of (4.2) has profound differences for the trend. If yp
is the initial condition, itis readily verified that the general solution to the first-order
difference equation represented by (4.3) is
yt ey Wécan tn of i sosend component ars sees ince ee
Utne of ny shcks entropy However each shook epcens sin
theimercopt Sac al var havea cout af ty. cea ea eck
UOhe a sochufe end since ach, shock input pest alle andom,
Shang inthe coral san of Sos rape 0s pe of dt sur 6