Table 3 reports the coefficient estimates of vector autoregres-sion with one lag for each endogenous variable. Clearly, past
hot money has a significant positive explanatory power on
stock returns in EMs because the t-statistic is 4.3238, much
larger than the critical value at 1-percent statistical significance
level. However, lagged stock returns have little forecasting
power for current hot money (the t-statistic is only 0.4778).
The following Granger causality tests can confirm the results.
GRANGER CAUSALITY TESTS
Some economic variables are significantly correlated with each
other but not necessarily meaningful. Causality is one of the
most difficult issues in finance and economics, as well as in
other social sciences. One possible way to deal with causality is
with Granger causality testing, which is a statistical measure of
causality based on prediction. A variable X1 “Granger causes”
another variable X2, if and only if the past values of X1 contain
information that helps predict X2 beyond the past values of X2
only (for details, please refer to Yan et al. 2016 or others).
Following the extant literature, I carry on Granger causality
tests to explore whether the variation of hot money in equity
flows (stock returns of EMs) does contribute to the change of
stock returns of EMs (hot money in equity flows).
According to the results shown in table 4, hot money in equity
flows only has unilateral Granger causality with stock returns in
EMs; that is, hot money is the Granger causality of stock
returns and stock returns are not the Granger causality of hot
money. In other words, changes of stock returns do not yield a
significant impact on the movement of hot money.
introduce VAR models is that they provide us with a general
method to evaluate bi-directional causality; that is, on the one
hand, hot money may cause stock prices to rise or fall, and on
the other hand, the stock returns would drive hot money flows.
The VAR I estimate can be modeled as
Yt=C+π1Yt- 1+π2Yt- 2+…+πk Yt-k+μt ( 1)
where μt~i.i.d.N (0,Ω).
I can display the VAR model in a compact form for t= 1, 2, …. T,
where Yt, C, and μt are 2× 1 column vectors, and πi is a 2× 2 coefficient matrix:
[ [[[ ]]] ] Hot Moneyt C1 μ1t π11,i π12,i Yt= , C= , μt= , πi= , i = 1, 2,...k Stock Returnt C2 μ2t π21,i π22,i
The unknown parameter C is the constant intercept term, πi is
the coefficient of the endogenous variables, and μt is the disturbance vector. I use aggregate monthly data for hot money and
equity returns across all emerging-market economies, covering
a sample period from 1993–2013. Hot Money is the temporary
component of equity flows from the United States to EMs
scaled by the local equity market capitalization; Stock Returns
are a monthly percentage of value-weighted returns on emerging stock indexes.
VAR MODEL COEFFICIENTS
I use the Akaike information criterion and Schwartz-Bayes
criteria to specify the appropriate lag length of the VAR model,
which turns out to be a lag length of one. The VAR model
employed eventually can be written as
 μ1t μ2t [ [ [ [ ] ] ] ] Hot Moneyt Hot Moneyt– 1 C1 π11,i π12,i = + + Stock Returnt Stock Returnt– 2 C2 π21,i π22,i ( 2)
OVERALL VAR ESTIMATES
table 3 reports the pooled coefficients estimates of one-month lagged hot money and stock returns for all the markets in my sample
in aggregate. Left (right) results are for one-month lagged hot money (stock returns). The numbers in the second row (in italics) are
t-statistics for the null hypothesis that the corresponding coefficient of hot money or stock returns is zero. the VAR coefficients and covariance matrix are estimated by ordinary least squares (oLS). the sampling frequency is monthly during January 1993–december 2013.
Hot Money Stock Returns
coefficient t−statistics coefficient t−statistics
Hot Money(− 1) 0.4261 7.2360 0.0142 4.3238
Stock Returns(− 1) 0.5253 0.4778 0.1527 2.4881
Intercept − 2.8044 −0.3513 0.3874 0.8693
OVERALL GRANGER CAUSALITY TESTS
table 4 reports F-statistics, P-values, and conclusions for the null hypothesis of ‘no Granger causality’ either from stock returns to hot
money, or from hot money to stock returns for all the markets in my sample in aggregate. The sampling frequency is monthly during
January 1993–december 2013.
Null Hypothesis F-Statistic P-value Conclusion
Stock Returns does not Granger Cause Hot Money 0.2282 0.6333 cannot reject
Hot Money does not Granger Cause Stock Returns 18.6951 2.E05 reject