Professional Documents
Culture Documents
Herman J. Bierens
Pennsylvania State University
Consider two independent unit root processes, ∆yt ' u t and ∆xt ' vt , where the ut's and
ut 0 1 0
- i.i.d. N2 , . (1)
vt 0 0 1
Then yt, with all leads and lags, is independent of xt, with all leads and lags. If we regress yt on xt
for t = 1,..,n, one would therefore expect the slope coefficient to be insignificant, but as we show
now, that is not true. First, consider the OLS regression of yt on xt without an intercept. The OLS
estimate of the slope is:
't'1xtyt
n n t t
(1/n)'t'1 x0/ n%(1/ n)'i'1vi y0/ n%(1/ n)'j'1uj
't'1xt
γ̂0 ' '
n 2 n t 2
(1/n)'t'1 x0/ n%(1/ n)'i'1vi
(2)
n
(1/n)'t'1 x0/ n%Wx,n(t/n) y0/ n%Wy,n(t/n)
' ,
2
(1/n)'t'1 x0/ n%Wx,n(t/n)
where
with [rn] the largest natural number # rn. Since these functions are step functions, and both
Wx,n(1) and Wy,n(1) are standard normally distributed, hence Wx,n(1) = Op(1) and Wy,n(1) = Op(1),
we have
1
n&1 t%1
(1/n)j Wx,n(t/n) ' (1/n)j Wx,n(z/n)2dz % Wx,n(1)2/n
n
t'0 m
2
t'1
t
(5)
n
m m
' (1/n) Wx,n(z/n)2dz % Wx,n(1)2/n ' Wx,n(r)2dr % Op(1/n) ,
0
and similarly
m
(6)
t'1
m
(7)
t'1
The integrals in (5) through (7), and in the sequel, are taken over the unit interval [0,1], unless
otherwise indicated.
It can be shown that the integrals in (5) through (7) converge jointly in distribution, due to
the functional central limit theorem and the continuous mapping theorem:
m x,n m m
T
LEMMA 1. W (r)2dr , Wy,n(r)2dr , Wx,n(r)Wy,n(r)dr converges in distribution to
m m m
T
Wx(r)2dr , Wy(r)2dr , Wx(r)Wydr , where Wx(r) and Wy(r) are independent standard Wiener
processes.
1
Billingsley, Patric: Convergence of Probability Measures, John Wiley, New York, 1968
2
in distribution. Note that the limiting random variable γ0 is continuously distributed, and in
particular, P( γ0 = 0) = 0.
Along the same lines, and using (8), it follows that the residual sum of squares, RSS0, of
the regression involved, divided by n2, satisfies
j j
RSS0 n n
m y,n m
1 2 2 1 2 2
' yt & γ̂0 xt ' W (r)2dr & γ̂0 Wx,n(r)2dr % op(1)
2
n n 2 t'1 n 2 t'1
(9)
m y m
2
6 W (r)2dr & γ0 Wx(r)2dr in distr.
n 2 n 2
tˆ0 γ̂0 (1/n)'t'1xt γ̂0 (1/n 2)'t'1xt
' ' n/(n&1)
n RRS0 / (n&1) RRS0 / n 2
(10)
m x,n m x
γ̂0 W (r)2dr γ0 W (r)2dr
' % op(1) 6
m y,n m x,n m y m
2 2
W (r) dr & 2
γ̂0 W (r) dr 2
W (r)2dr & γ0 Wx(r)2dr
Similar results as in (8) and (10) hold for slope parameter γ̂1 and corresponding t-value
γ̂1 6 γ1 ' m m m
Wy(r)Wx(r)dr & Wy(r)dr Wx(r)dr
(11)
m x m
2 2
W (r) dr & Wx(r)dr
and
m x m
2
tˆ1 W (r)2dr & Wx(r)dr
6 γ1 (12)
m y m m x m x
n 2 2 2 2
2
W (r) dr & Wy(r)dr & γ1 2
W (r) dr % γ1 W (r)dr
3
γ1 m m x
2
2
Wx(r)2dr & W (r)dr
R2 6 (13)
m m
2
Wy(r)2dr & Wy(r)dr
in distribution
The conclusion from these results is that one should be very cautious when conducting
standard econometric analysis using time series. If the time series involved are unit root
processes, naive application of regression analysis may yield nonsense results.