We saw in section 2 that the stochastic component of the return on the
hedge portfolio, y, corresponds to a displaced chi-squared distribution with
one degree of freedom. Although the return on the hedge portfolio will vary
both on account of changes in i and the stochastic nature of y, to focus
attention on the changes induced by the stochastic component we assume I
is held constant.3 As mentioned before, assuming homoscedasticity is not
valid, but homoscedasticity can, in principle, be achieved by varying the
number of options held inversely with 2. (Assuming, of course, that the
higher-order terms in dt remain negligible.)
In this section we analyse the distribution of hedge portfolio returns on
this basis. In particular, we examine the biases which can arise from the
skewness of the y-distribution in the calculation of t-statistics for confidence
interval construction. The results presented in this section are illustrative
rather than definitive since formula (IO) for HR neglects higher powers of t.
With 1. assumed constant, the estimation bias will exactly correspond to
that obtained in the following experiment. A random sample of size n is
taken from a chi-squared distribution with one degree of fredom and tstatistics
for confidence limits on the sample mean are computed. What is the
resulting distribution of the t-statistics like? How does the bias change as n
changes? We are assured that for large samples, the t-statistics will be
‘correct’ by an appeal to the law of large numbers. However, it is of interest
to examine the behavior of the c-statistics for small values of n such as 10, 20
and 30 since it has been suggested that weekly rebalancing of hedge
portfolios may be adequate and also since the 60 trading days of a threemonth
option is not necessarily large enough for the law of large numbers to
apply. For small sample sizes, the c-statistics will be biased downwards. This
is because a negative mean return is caused by a predominance of negative
daily hedge returns. These negative returns are closely bunched together and
the variance estimate tends therefore to be low. This leads to a large negative
r-statistic. Conversely a positive mean return is caused by a predominance of
positive widely dispersed daily returns. The t-statistic tends to appear
insignilicant because of the resulting high variance estimate.
Table 2 shows the result of a simulation experiment. The approximate
frequency distribution of t is displayed for the case when the parent
population is chi-squared and when it is normal.
As can be seen, there are substantial differences in the distribution of the
estimated t-statistic when the sample is taken from the chi-squared distri-
‘In the next section we analyse the structure of hedge returns when the option is held to
maturity, and we thus account for the combined effect of variations m I and in the stochastic
component. The distribution of hedge returns in this case will be examined using Monte Carlo
simulation and so the higher order (At) terms, ignored in sections 3 and 4 will be automatically
Included in the analysis.
8/3-