Applied Financial Beonometries Goh Kim Leng
(a) Runs test
Price changes can be positive, negative or zero depending on whether the
price in the current period has increased, declined or remained unchanged
from the previous period. A run is defined as a sequence of successive price
changes of the same sign. There are 3 types of run: plus, minus ot zero. The
actual number of runs R is then the sum of the number of runs for plus,
minus and zero price changes. If Hp: price changes are random is true, then
ECR) < (Mt - 2 mi [n ,
3 mt [3 m4 nlrtt)) -2n 4, mn
n? (a-1)
where n is the number of observations (price changes)
m) is the number of runs of each type (i=1 for positive price changes, i=2 for
negative price changes and +3 for no price changes).
Var (CR) =
For la
ge n, the test statistic
Rto.5 -E(R)
ie ql Var CR)
follows approximately a standard normal distribution under Hy.
rs NCO,1)
The discontinuity adjustment is
40,5 if REE(R) and -0.5 if R>E(R).
‘The rejection region is Z<-Z,, fora a-level test.
Advantage of the test is that it does not require any assumption about the
underlying distribution (non-parametric). Weakness is that the test examines
only one-period changes.
(b) Autocorrelation Coefficients
Serial dependence may be found over short periods of time (more than one
period). The autocorrelation coefficient of lag k provides a measure of theApplied Financial Econometrics Gob Kim Leng
strength of the linear relationship between the value of a variable at time t
and its value k periods earlier. The autocorrelation coefficient of lag k is
given by 2
~ 2 Ty. -y)
(ve-FVe-n -V)
eck
bs 2.
aie
z, (Ver?
4
where, ori =
7 ako
If the n is small, we should use the adjusted autocorrelation coefficien
fis. ait
Va * Wk *
Note that n, is computed in Eviews (not the adjusted r,), The null hypothesis
ig Ho: py =0 where py is the population autocorrelation coefficient of lag k.
If Hy is true, then the statistic
, \
hv N¢ “y ae)
4. , ——
The test statistic is given by ds Vp |W ke
For Hy: p= 0, the rejection region is given by |Z [Sze
(c) Liung-Box Q-Test
Consider hypothesis testing at significance level . To test for serial
correlations at different lags, several separate tests must be performed. The
significance level of the combination of tests is quite different from a. To
test for
Ho! fr = Pa = ++ = Pls
we use the following statistic:
taApplied Financial Beonometries Gob Kim Leng
(nea) ee SH) Uidite th, eC
Silsh viconas
The rejection region is: Qi > Zeke
stic is computed in Eviews.
(d) von Neumann’s Ratio Test
This test can be used to test for serial correlation in the return series v;,
Vay s+ Vae The test statistic is given by:
Ly,
Bley) *[On- 1)
x
ra (0) Tn
Under Hy: no serial correlation in returns or p= 0, V follows a normal
distribution with mean and variance:
Zyx where Z= ——
{var W)
Note that the tests in (b), (c) and (d) are large-sample tests. In small samples,
the assumption of normality is crucial.