Professional Documents
Culture Documents
The Causes of Stock Market Volatility in Australia: Applied Financial Economics, 1998, 8, 597-605
The Causes of Stock Market Volatility in Australia: Applied Financial Economics, 1998, 8, 597-605
The paper examines the extent to which the conditional volatility of stock market
returns in a small, internationally integrated stock market are related to the condi-
tional volatility of financial and business cycle variables. It employs a low frequency
monthly dataset for Australia including stock market returns, interest rates, inflation,
the money supply, industrial production and the current account deficit over the
period from July 1972 to January 1994. A novel feature of the analysis is the
estimation strategy employed to overcome the generated regressors problem which
pervades some recent related research. Specifically, the procedure of employing
a two-stage estimation process to first estimate conditional volatilities and then model
their interrelationships yields inefficient estimates, introduces bias into a number of
diagnostic test statistics and generates potentially invalid inferences. This problem is
overcome in the current paper by jointly estimating the equation for the conditional
volatility of stock market returns together with the equations determining the condi-
tional volatilities of all variables included in the model using the generalized least
squares (GLS) estimation procedure together with the Hendry general-to-specific
modelling strategy. Among the most important determinants of the conditional
volatility of the Australian stock market are found to be the conditional volatilities of
inflation and interest rates which are directly associated with stock market volatility,
and the conditional volatilities of industrial production, the current account deficit
and the money supply which are indirectly associated with stock market conditional
volatility. Among these variables, the strongest effect is found to be from the condi-
tional volatility of the money supply to the conditional volatility of the stock market.
By contrast, no evidence is found of volatility spillover from the foreign exchange
market to the stock market in Australia.
I. INTRO DUCTI ON leverage, Merton (1980), Poterba and Summers (1986) and
French et al. (1987) relate stock market volatility to the
The issue of stock market volatility has received much volatility of expected returns, and Schwert (1989) conducts
attention in the finance literature. The main questions which an extensive array of tests on the macroeconomic causes of
have been addressed include: what are the important causes stock market volatility over long runs of monthly data for
of stock market volatility? has it increased over time? has the United States. The issue of whether the world’s financial
international financial integration led to faster transmission and capital markets are now transmitting volatility more
of volatility across international stock markets? and what quickly has been examined by Koch and Koch (1991),
role, if any, ought regulators play in the process? Previous Malliaris and Urrutia (1992), Chan et al. (1992) and
researchers have examined these issues. For example, on the Rahman and Yung (1994). The related issue of whether
causes of stock market volatility, Officer (1973) examines the stock market volatility has increased over time has
effects of volatility in business cycle variables, Black (1976) been investigated by Peel et al. (1993), while Scott (1991)
and Christie (1982) relate stock market volatility to financial and Timmermann (1993) examine the extent to which
0960—3107 1998 Routledge 597
598 C. Kearney and K. Daly
the volatility of stock prices determines their underlying research by Koutoulas and Kryzanowski (1996) employs
value. a similar methodology to that which is employed in this
The purpose of this paper is to contribute to the literature paper in order to examine the explanatory power of macro-
on the causes of stock market volatility by examining the economic conditional volatilities in an arbitrage pricing
determinants of movements in the volatility of equity re- model of monthly Canadian stock market returns. Interest-
turns in a small, internationally integrated stock market. ingly, they find that the conditional volatilities of industrial
The Australian stock market serves as an especially illumi- production and the exchange rate feature in their model
nating example in this regard insofar as it is increasingly while interest rates play a more ambiguous role.
integrated with the important international stock markets The paper is structured as follows. The next section de-
throughout Asia and the world. Previous research on the scribes the theoretical framework which motivates the em-
volatility of the Australian stock market includes the work pirical analysis and describes the dataset which is used for
of Brailsford and Faff (1993) and Kearns and Pagan (1993), the study. Section III turns the focus of attention to the
both of whom examined the relative explanatory power causes of volatility in the Australian stock market. It em-
of alternative models of conditional volatility, but neither of ploys the generalized least squares (GLS) estimation meth-
whom related stock market volatility to the volatility of odology to implement the Davidian and Carroll (1987)
financial and business cycle variables. This paper develops ARCH model of conditional volatility to model the deter-
and estimates a model which is capable of explaining the mination of Australian stock market conditional volatility
financial and business cycle determinants of movements in using the Hendry general-to-specific estimation methodo-
the conditional volatility of the Australian All Industrials logy (see Mizon (1995) for a recent overview). The final
stock market index. It employs a low frequency monthly section reviews the main results and draws together the
dataset including stock market returns, interest rates, infla- conclusions.
tion, the money supply, industrial production and the cur-
rent account deficit over the period July 1972 to January
1994. Among the most important determinants of Austra- II . TH EORY, METHO DOLOGY AND DATA
lian stock market volatility are found to be the volatility of
inflation and interest rates which are directly associated The price of equity at any point in time is equal to the
with stock market volatility, while the volatility of industrial discounted present value of expected future cash flows (in-
production, the current account deficit and the money cluding capital gains and dividends) to shareholders:
supply are indirectly associated with stock market volatility.
Among these variables, the most significant determinant of CG (IP , M , P , D , S )
conditional volatility in the Australian stock market is the E QG"E R>I R>I R>I R>I R>I R>I
R\ R R\ (1#R )I
degree of volatility which exists in the money supply. By I R>I
(1)
contrast, no evidence is found of a statistically significant
relationship between foreign exchange market volatility and where QG is the price of asset i, CG denotes the cash flows
stock market volatility in Australia. R
associated with it, R denotes the interest rate and E is the
The results presented in this paper constitute new evid- R\
expectations operator. In an open economic and financial
ence which is interpretable as an extension of the low system such as that which prevails in Australia, corporate
frequency analysis of Schwert (1989) who did not include cash flows CG are influenced by developments in the macro-
international factors such as the current account deficit and economy such as the level of aggregate industrial produc-
the exchange rate in his investigation of the causes of stock tion (IP), the money supply (M) and the level of prices (P),
market volatility in the United States. In addition, Schwert’s the external trading position of the country as measured by
(1989) vector autoregressive model of the determination of the current account deficit of the balance of payments (D)
conditional stock market volatility, along with other re- and the spot exchange rate (S) which is defined as the
search which employs the same methodology, suffers from domestic currency price of foreign exchange.
the generated regressors problem (Pagan, 1984, 1986; In moving from equity prices to returns, let qG denote the
R
McAleer and McKenzie, 1991; Oxley and McAleer, 1993) actual return on asset i, and let the q' G"E [qG/I ] denote
which yields inefficient estimates, introduces bias into R R R R\
its expected return conditional on the available information
a number of diagnostic test statistics and generates poten- set at time t!1. In addition, let pqG denote the uncondi-
tially invalid inferences. This problem is overcome in the R
tional standard deviation of returns on asset i, and let
current paper by jointly estimating the equation for the pL "E [p /I ] denote its conditional counterpart. From
qG qG
Q: Australian sharemarket All Industrial Index, sampled on close of trading on the last trading day in the month. The index is re-based to
1973 " 100. Source is Australian Stock Exchange.
IP: The monthly percentage change in the index of industrial production, rebased to 1973 " 100. Source is OECD Economic Database.
P: The monthly index of wholesale prices in Australia, re-based to 1973 " 100. Source is OECD Economic Database.
I: The monthly percentage change in the wholesale price index (P) in Australia, re-based to 1975 " 100. Source is OECD Economic
Database.
D: The current account deficit of the database of payments in Australia, in billions of Australian dollars. Source is OECD Economic
Database.
S: The spot exchange rate of the Australian—US dollar exchange rate, sampled at the end of the last trading day each month. Source is
OECD Economic Database.
R: The interest rate on 3-month bank accepted bills in Australia, sampled at the end of the last trading day each month. Source is Reserve
Source Bank of Australia.
CRASH: Dummy variable to capture the effects of the October 1987 stock market crash. It takes the value of 1 in October 1987 and 0 at
all other times.
SD: Monthly seasonal dummy variables to capture the effects of monthly variations the conditional mean and conditional standard
deviations of the business cycle and financial variables.
600 C. Kearney and K. Daly
II I. MOD ELLI NG TH E C ONDITI ON AL In the current context of estimating the conditional vola-
V OL AT I L IT Y O F TH E A U ST R A LI AN tilities of financial and economic variables, the potential
ST OCK MA R KET for efficiency gain by estimating the system by GLS is
considerable.
In order to appropriately account for the non-zero cross- Table 2 presents the summary statistics from the OLS
equation covariances which arise from the generated estimates of the first 12 autoregressions contained in Equa-
regressors problem (Pagan, 1984, 1986; McAleer and tions 4 and 5 which are used as inputs during the GLS
McKenzie, 1991; Oxley and McAleer, 1993), the equation estimation procedure. Looking first at the top part of the
for the conditional volatility of the Australian stock market table which summarizes the equations for the conditional
index is estimated jointly with the equations determining means, the equations explain between 7 and 79% of the
the conditional volatilities of all variables included in the variation in the dependent variables, with, as expected,
model using the generalized least squares (GLS) estimation the financial variables (including the stock market returns,
procedure. The GLS estimator is described by first stacking the interest rate and the exchange rate) occupying the lower
the system to be estimated in the following form with end of the explanatory power rankings. The equations are
M equations and ¹ observations: all free from higher order autocorrelation as evidenced by
the Ljung—Box Q-statistics. The sums of the lagged depen-
y X 2 0 b e dent coefficients (with their marginal significance levels in
" # (6)
y 0 2X b e brackets) are presented in the fourth column of Table 2
K K K K
under the heading Sum. The columns headed by F and
where y is an M¹ ; 1 vector of dependent variables, X is an
F provide the F-statistics for the joint exclusion of, respec-
M¹ ; ( + K ) matrix of explanatory variables, b is an
G G tively, all lagged dependants and all the seasonal variables
M¹ ; ( + K ) matrix of coefficients and e is an M¹ ; 1
G G in each regression. The lagged dependants in the equations
vector of errors. In the model analysed here Equation 6 con- for the stock market, inflation and the current account
tains 15 equations (i.e. M " 15) with 259 observations (i.e., deficit are jointly significant in their respective equations,
¹ " 259) on each variable. The first seven equations are while the seasonal dummy variables are jointly significant in
given by Equation 5 for each variable in the X vector, the all equations except those for inflation and the exchange
next seven equations are given by Equation 4 for each rate. With regard to the Australian stock market, the latter
variable in the X vector, and the last equation is given by finding is in conformity with other researchers such as
p̂/"j #j (K) pL V#e (7) Gultekin and Gultekin (1983) who found statistically signifi-
R R R cant stock market seasonality in 14 out of 17 countries
where the X vector is X(Q, IP, M, P, D, S, R) as before, and studied, Kramer (1994) who related stock market seasonal-
the j (K) are polynomials of the 4th degree in the lag ity in the United States to underlying macroeconomic sea-
sonality, and Haugen and Jorian (1996) who have reported
operator, K. Equation 7 relates the conditional volatility of
the Australian stock market, Q, to the conditional volatility that the January effect has not declined in size during the
of the Australian financial and business cycle variables in- past three decades in the United States.
cluding industrial production IP, the money supply M, Looking next at the bottom part of Table 2 which
inflation P, the current account deficit D, the exchange rate summarizes the equations for the conditional standard
S, and the interest rate R. deviations, the equations explain predictably less of the
The system in Equation 6 incorporates the possibility of variation in the dependent variables, and all are free
cross-equation correlation among the error terms which from higher order autocorrelation as evidenced by the
yields the following covariance matrix of errors: Ljung—Box Q-statistics. The lagged dependants in the equa-
tions for the stock market, industrial production, inflation,
p I p I the current account deficit and the interest rate are jointly
E(ee)")" %
p I p I significant in their respective equations, while the seasonal
% %%
dummy variables are, as in the top part of Table 2, jointly
Writing the system in matrix form as in Equation 8, we can significant in the equations for all variables except inflation
describe the GLS estimator as given by Equation 9: and the exchange rate.
Figure 1 presents time series plots of the conditional
y"Xb#e (8)
volatilities of all variables used in the study. Most series
b"(X)\X)\ X)\y (9) display considerable time variation in their volatilities. Of
particular note is the high volatility of interest rates during
The gain in efficiency of this GLS estimator over OLS the 1980s and the stock market volatility associated with the
estimation depends on a number of factors including the October 1987 crash. The summary statistics of each series
extent of relations across the equations in the system and are presented in Table 3. The means of the variables show
the extent to which the X variables differ across equations. considerable variation, ranging from a high of 0.089 for the
Causes of stock market volatility in Australia 601
Table 2. Estimation results of the ARCH models of stock market returns and business cycle variables July 1970—January 1994
The table presents the regression summary results for Equations 2 and 3 in the text, regressed on each variable in the table. SEE denotes the
standard error of the regression and Q denotes the Ljung—Box test for higher order autocorrelation. Sum denotes the sums of coefficients of
the lagged dependent variables in each regression equation. F denotes the F-statistics for the joint exclusion of all lagged dependants in
each regression. F denotes the F-statistics for the joint exclusion of all the seasonal coefficients in each regression. Marginal significance
levels for these test statistics are presented in the brackets.
current account deficit (D) to a low of 0.010 for the money this analysis, the model was estimated by GLS along the
supply (M), with the intermediate rankings being in order lines of the general-to-specific modelling strategy (Mizon,
from highest to lowest, interest rate (R), industrial produc- 1995). This involved first estimating the system described in
tion (IP), stock market (Q) and inflation (P). These rankings Equation 6 with the general specified lag structure for the
are also mostly replicated in terms of the variances. There is conditional volatility of the stock market described in Equa-
evidence of non-normality as indicated by the marginal tion 7, and then sequentially restricting the latter equation
significance levels of the skewness of most variables and of by excluding its statistically insignificant components. The
the Kurtosis of some variables. The two right-hand columns final form of the equation for the conditional volatility of
of Table 3 provide the minimum and maximum values of the stock market is presented in Equation 10 and the
the conditional volatilities along with their dates in estimation results are presented in Table 5.
brackets. As expected, the stock market crash features as
being the most volatile month during the study period for p̂/"j #j Crash#j p̂/ #j p̂/ #j p̂'.
R R\ R\ R\
the stock market variable, and the early and mid-1980s
#j p̂+#j p̂.#j p̂. #j p̂"
feature as being the period of most volatile interest rates. R R R\ R\
As a precursor to the final model specification, the gener- #j p̂" #j p̂##j p̂0 #e (10)
ated conditional volatilities were tested for their order of R\ R R\ R
integration in order to avoid any possible spurious regres- Looking first at the model’s overall performance as present-
sion relationships. Although inspection of Fig. 1 suggests ed by the equation diagnostics in the bottom part of Table 5,
that the variables are all stationary, it is advisable to check the R statistic indicates that it explains an average of 35%
the correctness of this supposition. Given the non-normality of the variation in the conditional variance of the Australian
of some variables, Table 4 presents the results of unit stock market, with the standard error of the estimate being
root tests for each variables using alternatively the equal to one-third of the mean of the dependent variable.
Dickey—Fuller (1979) and the Phillips—Perron (1988) testing This compares well with related work of, for example, Stew-
methodologies. As expected, each test strongly indicates ard (1993). The regression F-statistic is highly significant,
stationarity of every variable employed in the model. indicating that the included variables are jointly statistically
Since financial theory does not contain predictions about significant determinants of the conditional volatility of the
the form of the lag structure which may be appropriate in Australian stock market. The Durbin—Watson (D¼) and
602 C. Kearney and K. Daly
included in Table 5; CHO¼ splits the sample in half
(from 1974:1—1984:12 and from 1985:1—1994:1), CHO¼
splits it into 1/3 and 2/3 (from 1974:1—1981:12 and from
1982:1—1994:1), while CHO¼ splits it into 2/3 and 1/3
(from 1974:1—1987:12 and from 1988:1—1994:1), respectively.
These test statistics confirm that the model’s structure re-
mains stable over time.
Looking next at the individual coefficient estimates of the
lagged dependent variable, the j and j coefficients indi-
cate that the model incorporates significant lagged depen-
dent terms which implies that volatility in the Australian
stock market responds dynamically to variations in the
volatility of the financial and business cycle variables. The
equation also contains a positively signed and statistically
significant constant term with a value of j " 0.035 and
associated t-statistic of 6.07 which indicates that stock mar-
ket volatility in Australia tends to proceed independently
from the other influences included in the model. The dummy
variable for the October 1987 stock market crash is
also positively signed as expected with a coefficient of
j " 0.063, and it is strongly statistically significant with
a t-statistic of 5.89.
Turning now to examine the effects of the conditional
volatility of the business cycle variables on the conditional
volatility of the Australian stock market, all variables are
statistically significant at the 5% level of significance except
for the conditional volatility of the exchange rate. Increases
in the conditional volatility of interest rates and inflation are
associated with higher stock market conditional volatility,
while increases in the conditional volatility of industrial
production, the current account deficit and the money
supply are associated with lower stock market conditional
volatility. More specifically, the positive j coefficient of
0.078 (with its t-statistic of 2.42) indicates that the condi-
tional volatility of interest rates impacts positively upon the
conditional volatility of the stock market with a one-month
lag. The negative j coefficient of !0.541 (with its t-statistic
of 2.00) and the positive j coefficient of 0.853 (with its
t-statistic of 3.09) together imply that the conditional vola-
tility of inflation has a net positive impact on the condi-
tional volatility of the stock market with a one-month lag.
By way of contrast, the negative j coefficient of !0.088
(with t-statistic of 3.11) indicates that the conditional volatil-
ity of the stock market varies indirectly with the conditional
Fig. 1. Conditional volatilities of financial and business cycle vari- volatility of industrial production with a three-month lag.
As the negative j coefficient of !1.093 (with its t-statistic
ables
of 4.41) indicates, the same qualitative relationship pertains
to the effect of changes in the conditional volatility of the
money supply on the conditional volatility of the stock
Kolmogorov—Smirnov (K!S) statistics indicate that the market, although in this case the strength of the relationship
model is free from first and higher order autocorrelation. is greater and the impact is immediate (i.e. within one
The ARCH test for heteroscedasticity indicates that this is month). Indeed, the magnitude of the j coefficient together
not a problem, which is intuitive insofar as we are working with its associated t-statistic indicates that, apart from the
with the conditional volatilities of all variables. Finally, stock market’s recent volatility history and the dummy
three versions of the Chow test for structural stability are variable for the October 1987 stock market crash, the most
Causes of stock market volatility in Australia 603
Table 3. Descriptive statistics for the conditional volatilities of variables used in the analysis
Minimum Maximum
Variable Mean Variance Skewness Kurtosis value value
Q 0.033 (0.00) 2.01\ 1.68 (0.00) 6.78 (0.00) 0.002 (88 : 8) 0.121 (87 : 12)
½ 0.039 (0.00) 6.66\ 1.48 (0.00) 3.32 (0.00) 0.001 (90 : 3) 0.164 (80 : 6)
M 0.010 (0.00) 9.23\ 0.50 (0.00) 0.06 (0.85) 0.002 (82 : 2) 0.018 (91 : 1)
P 0.011 (0.00) 9.19\ 0.62 (0.00) 0.37 (0.24) 0.005 (76 : 11) 0.023 (86 : 9)
D 0.089 (0.00) 9.40\ 0.79 (0.00) 0.50 (0.11) 0.035 (77 : 10) 0.201 (93 : 12)
E 0.019 (0.00) 4.21\ 0.52 (0.00) 0.45 (0.16) 0.002 (77 : 3) 0.040 (77 : 1)
R 0.053 (0.00) 5.55\ 0.84 (0.00) 0.99 (0.00) 0.012 (85 : 6) 0.148 (83 : 4)
The descriptive statistics are all calculated over the period 1973:8—1994:1. The figures in parentheses following the mean, skewness and
kurtosis are the marginal significance levels for zero. The figures in parentheses following the minimum and maximum values provide the
dates.
Table 4. ºnit root tests of stock markets and business cycle Table 5. G¸S estimation of the ARCH model of conditional volatil-
variables ity of the Australian stock market