Introduction

Traditionally, the hedge ratio is assumed to be constant over the relevant time. However,
recent empirical financial research reveals that asset returns volatility is time-varying, Park
and Switzer (1995) suggest that consistent hedge ratios may not be appropriate as the joint
distribution of cash and futures prices is changing over time, while Baillie and Myers (1991)
also prove that the conditional distribution between cash and futures price changes then
hedge ratios will vary over time. To obtain time-varying hedge ratios, accurate volatility
forecasting is necessary. Volatility is directly related to the uncertainty and risk of the market
and is the most concise and effective indicator to reflect the quality and efficiency of the
financial market. Thus, it is of great importance in the fields of securities investment, asset
pricing and risk management. Volatility forecasting models can be used to derive hedged
ratios. Incorporating time-variation into the hedge ratio improves the performance of the
hedge in terms of risk reduction (McMillan, 2005). This paper supports the view that an
accurate estimation of the volatility process plays an essential role in the construction of
hedging strategies and improving the hedging effectiveness.

The main task of this report is to act as a risk manager for a large American company that has
a target of hedging a long position in the S & P 500 index with a short position in the FTSE
100 index. The hedged portfolio is expected to perform better than the unhedged portfolio in
terms of risk. Risk characteristics are presented by variance in this paper. To construct this
hedge portfolio, it is necessary to estimate the time-varying minimum-variance hedge ratios,
which can improve the risk-reduction capacity better than the constant hedged ratio. This
paper adopts the rolling window model and the EWMA model to obtain estimates needed for
the time-varying minimum-variance hedge ratios.

The motivation for the analysis is to examine the important role volatility forecasting plays in
the construction of the hedged portfolio and how to use different models to estimate volatility
to get time-varying minimum-variance hedge ratios. The accurate volatility forecasting is
important not only because the unobservable variance of returns make it difficult to make an
investment decision, but also because the effects of volatility can be offset through hedging.
Hence, accurate estimation of volatility is of value in efficiently and effectively implementing.
Since volatility tends to aggregate, the natural way to estimate time-varying volatility is to
use estimates based on recent returns. Through the utilizing of the proper models, the change
of volatility could be calculated with more precision and accuracy (Kumar,2006). However,
there is no definite answer to the question of which model should be used to estimate the
volatility. In order to predict volatility efficiently and practically, this paper mainly uses the
rolling window model and the EWMA model to estimate the volatility. Therefore, we use
these two models and compute the optimal hedge ratio. The methodologies used in this
research are discussed in detail in terms of their principles, features, advantages and
disadvantages in the part of the method. Also, the efficiency of our hedge ratios is verified by
the reduction of Variance after using hedging strategies.

The main findings are as follow: we summarize the features of the sample, which can provide
overall stock performance in S & P500 index and FPSE 100 index. Then, we obtain hedge
ratios by implementing two models to predict volatility. We find that using hedging a long
position in S & P500 index with a short position in FPSE 100 index could effectively reduce
the variance of returns. This is due to the fact that variance reflects the magnitude of volatility.
The reduction in volatility indicates a reduction in risk (Wang, 2015). Last but not least,
different volatility estimates resulting from these two models, and the main findings are
showed in the following section of results.

The outline of this report is as follows: Section two introduces the principal, process of data
selection and summary statistics. Section three describes the theories and methods used in the
practical analysis of this article. With data obtained from online recourse, the rolling window
and EWMA volatility forecasting models are utilized in the process of estimating hedge ratios.
In the next section, we draw the results of the two models' volatility predictions and the
hedged portfolio we find. And in the last part, we summarized the research findings and
suggestions for future study.

In summary, this paper estimates the time-varying minimum-variance hedge ratio through
two models and examines the capacity of risk reduction through hedging strategies
constructed by two volatility-forecasting models.

Data
The data file used in our report was collected from Yahoo Finance
(https://uk.finance.yahoo.com/). It contains the historical data of the S&P 500 index and the
FTSE 100 index for the last six years, from April 1st, 2014 to March 31st, 2020. Observations
related to public holidays either in the United Kingdom and the United States was deleted, to
ensure the continuity and consistency of the data for the two indexes, leaving 1488 trading
days in our data set. Through the adjusted close prices of these 1488 observations, we
calculated the daily simple returns for these two indices over the six years. Summary
statistics for the data are given in Table 1.
Table 1 Summary statistics
^GSPC ^FTSE
MEAN 0.027% -0.006%
VARIANCE 1.21× 10$% 1.02× 10$%
SKEWNESS -0.793 -0.950
KURTOSIS 24.636 17.099
JB STATISTICS 37759.456** 12540.452**
Notes: The Bera-Jarque statistics is distributed asymptotically as a χ' (2) under the null of
normality. * and ** indicate significance at the 5% and 1% levels respectively.

It is evident that the GSPC returns series is slightly more volatile than FTSE. GSPC performs
better than FTSE in terms of mean returns in this period. Besides, both share indexes are
strongly non-normal as their JB statistics are extremely high. Also, they are leptokurtic and
significantly skewed to the left.

Method
In this part, the definition of volatility and the reason why there is a need for estimating
volatility is explained in detail. According to the case study, Volatility is defined as the
variability of security returns, and normally is measured by variance. In practice, the variance
of return is unobserved. Therefore there is a need for estimating volatility. In addition,
because of the assumption that the true variance is constant over time, we are able to estimate
variance by using the sample variance of past returns.

There are mainly two widely used methodologies in this research to estimate volatility;
Rolling window method and Exponentially weighted moving average model (EWMA). In the
following paragraph, the principles and characteristics of the two methods will be introduced,
and their limitations and advantages would also be compared and discussed.

The first approach is rolling window or moving average volatility model, which is the sample
variance estimated. It is based on the principle that time-varying volatility can be estimated
based on returns from the recent past, as volatility is normally clustered. The general form of
the model is as follows:

2
1 M -1
sˆ 2
t +2 = å (rt -i - rt )
M - 1 i =0
M is the length of the time period of the obtained data, which is also named as window length.
To get the optimal window length which can bring maximum risk reduction, data table is
used in EXCEL to repeat the process hundreds of times with different values of M, and find
the optimal window length. The advantages of this estimator are that it is very simple to
calculate and that except for the window size it does not involve any kind of estimation
(Suganuma, 2000). The window length is one of the most essential factors of the rolling
window model, it literally means how long the period of the date we utilise to forecast future
volatility. This factor is of great significance in the process of estimation, and two
considerations should be balanced in terms of choice of window length. From one perspective,
a long enough time period is needed to make sure the precision with which we estimate the
return variance. From the other perspective, the time period or window length should be short
enough to make the estimation to be more relative with the recent past data. This is because
of the fact that the everyday returns have the same amount of influence on the estimated
volatility in rolling window model. If the window length is too long, the impact of data from
a fairly long period of time ago could be overestimated. Moreover, according to the practice
experience, decreasing the window length increases the sensitivity of the rolling variance
estimator to observations that lie within the window, and consequently increases the volatility
of the volatility estimator. Therefore, in principle, rolling window model includes only the
most recent observations to estimate conditional volatility.

A noticeable drawback of rolling window model is ghost feature, which means that the lag
out of the observation window poses an impact on the estimates. From the fundamental point
of view, rolling window approach gives past observations either full weight (if they are recent
enough to lie within the window) or zero weight (if they fall out of the window). It may to
some extent influence the accuracy of the estimates. By comparison, exponentially weighted
moving average (EWMA) estimate can effectively overcome this problem.

The EWMA estimator yields a dynamic process for the conditional variance that is not
stationary, EWMA estimators regarding to the variance of a set of returns and the covariance
between two sets of returns can be written as
 2
σ̂ t+1 = λσ t2 + (1− λ )rt2
2 2
σ̂ ij,t+1 = λσ ij,t + (1− λ )ri,t rj,t
Because of the fact that short horizon returns in practice could normally be zero, we can
assume that the mean return is equal to zero. Under this assumption, the principle behind this
expression is to weight observations in the sample in a such way that their importance
declines smoothly into the past. According to Suganuma, in this method, the weights are
geometrically declining, so the most recent observation has more weight compared to older
ones. Therefore, this weighting scheme helps to capture the dynamic properties of the data.

The prerequisite for using EWMA estimator is that s 1 should be specified and so the
2

estimated variances are not unique; however, because of the exponential weighting, the

choice of s 1 rapidly becomes irrelevant as we move through the sample. For instance, we
2

could set s 1 to zero. Correspondingly, it is common to ignore the first 100 variance
2

estimates, so as to remove the effect of our specification of s 1 .

2

As the only unknown parameter of the function, decay factor λ can be set as the value that
optimises a particular economic or statistical criterion. In former research in 2000, Bollen
found that the optimal value of lambda is time varying and should be based upon recent
historical data. In financial practice, λ value of daily returns is usually estimated to be 0.94,
and for monthly return is 0.97. This result is proved to be infinitely close to the real cases and
therefore is reliable. This result is proved to be infinitely close to the real cases and therefore
is reliable. In our analysis, we first set it to 0.94, and then calculated the hedging effect under
this value. In addition, we calculated the value of λ under the optimal hedging effect through
the solver.

EWMA model is nested by Generalised autoregressive conditional heteroscedasticity

(GARCH) model. As a special case of GARCH, EWMA is more widely used in practice than
GARCH. This is due to the fact that GARCH model has two more parameters that are
difficult to estimate in reality, although EWMA replaces them with two estimated constant,
the results are proved to be Infinitely close to the real cases. In other words, EWMA model
could achieve near perfect results more efficiently.

In the case study, after obtaining the variance by meanings of Rolling window or EWMA
models, it is not difficult to calculate the time varying hedge ratio of the portfolio. The return
in hedging a long position in the S & P 500 index with a short position in the FTSE 100 index
is given by

+,,. = +0123,. − ℎ. +6718,.

where ℎ. is the time-varying hedge ratio. Thus, it is necessary to estimate the time-varying
minimum-variance hedge ratio, given by

90123,6718,.
ℎ. = '
96718,.

Thus, we need to estimate the conditional variance of FTSE 100 index returns, and the
conditional covariance between S & P 500 index returns and FTSE 100 index returns.
Meanwhile, by meanings of estimating the reduction of variance, we could also estimate the
time varying hedge ratio constructed by the two models. By doing so, we are able to measure
the effectiveness of our rolling window and the EWMA hedge ratio, by comparing the
difference of variance between unhedged portfolio and hedged portfolio. Thereby, we can
adjust the proportion of investment in a timely and correct way.

The research process contains the following preconditions. In the EWMA model, the size of
the hedge ratio is calculated in two cases that the delay factor equal to 0.94 and not equal to
0.94, so as to derive the change of volatility. The rolling window model calculates the size of
the hedge ratio and changes in volatility when the calculation window is 250 days. When the
rolling window is changed, the risk reduction capacity of hedged portfolios will change as
well.

In addition to the two models used in the case study, implied volatility also worth mentioning.
It is usually used in option pricing and is different from the former estimators in principle. As
the Rolling Window estimator and EWMA estimator are all backward looking, but the
Implied volatility estimating is forward forecast. Under the efficient market hypothesis, the
fair value of an option should equal to its market price, Given certain assumptions, option
price is given by a function of risk free rate, time to maturity, strike price and volatility of the
underlying asset price. In this case, all the other three parameters are known, and the fair
value of option is equal to the market price, which is also known, By inverting the
appropriate option pricing formula, we can infer the volatility of the asset price that is
consistent with the observed option price.
Results

This report selects the rolling window model and EWMA model, as mentioned before, to
analyze the possibility of reducing volatility through hedging a long position in S&P500
index with a short position in FPSE 100 index and, more specifically, estimate how much
reduction the hedging portfolio does. The results of these two different models are as follow:

Considering the rolling window model, the window length this research chooses is 250 days,
in order to contain as much data or situation as possible to increase the accuracy of estimating
return variance. After calculating simple returns of S&P 500 index and FPSE 100 index from
April 1st 2014 to March 30th 2020, a series of conditional variance and conditional covariance
for two index has been estimated. Then, it is accessible to construct the time-varying hedge
ratio and calculate hedge portfolio return and its variance which is an essential metric to
evaluate the feasibility of this hedge. Therefore, through model operation, the variance of
returns of S&P 500 index is at 0.013367%, while the variance of returns of S&P 500 index
after hedging by FPSE 100 index is at 0.007903%. Comparing the magnitude of hedge and
unhedged variance, the percentage of volatility of S&P 500 index has been reduced at 40.88%
across the hedge portfolio with 250 days window length. This implies that the hedge portfolio
has a positive impact on reducing the risk of investing in S&P 500 index alone.

Additionally, in order to calculate the optimal window length, the Data table is involved as an
indispensable data analysis tool in the analysis. It repeats the process of estimation with a
wide range of window lengths between 10 and 250 days and plots all the percentage of
declining of the variance of returns in figure 1. We find that, as a measure of hedging
effectiveness, the average risk reduction when using hedging strategies instead of no hedging
at all is almost stable at 40.00% (minimum 34.00%; maximum 40.90%).

Hedging effectiveness
50.00%
40.00%
reduction

30.00%
20.00%
10.00%
0.00%
0 50 100 150 200 250 300
window length

Figure 1 Hedging effectiveness

Comparing these results to it that is estimated with 250 days window length, we find 246
days window length is the most effective one. In this situation, we compare the variance of
returns of hedged portfolio and S&P 500 index, and it peaks at 40.90%, which illustrates the
greatest degree of decreasing volatility of S&P 500 index using the rolling window model.

The other model is EWMA estimator which uses an alternative way to weight observation.
This way is introducing a new variable defined as decay factor,l, to weight observation. At
the beginning of this approach, 0.94 is selected as a decay factor and s12 is set to 0. Then,
based on historical data from April 1st 2014 to March 30th 2020, we calculate simple returns
of S&P 500 index and FPSE 100 index. After that, this study estimates a series of conditional
variance and conditional covariance about two indexes across EWMA approach with s12 of 0
and l of 0.94 and calculates a series of time-varying hedge ratios according to these two
variables estimated before. Finally, using the time-varying hedge ratio constructed, the
variance of returns of S&P 500 index without hedging is at 0.0133372% and the variance of
returns of S&P 500 index after hedging by FPSE 100 index is at 0. 0081801%. Comparing
these two variance using same method as rolling window model previously, the proportion of
reduction of the volatility of S&P 500 index hedged by FPSE 100 index is at 38.67%. This
means that the hedge portfolio is efficient to reduce the volatility of S&P 500 index, which is
consistent with results from the former model. In addition, although the decay factor is set to
0.94, it is not the optimal one. 0.9905 is the excellent decay factor through the solver
approach. When l is equal to 0.9905 and other conditions remain consistent, the maximum
percentage of reduction of the variance of returns of S&P 500 index hedged by FPSE 100
index is at 40.68%. Therefore, under EWMA estimator, the maximum volatility that a hedged
portfolio can reduce is 40.68%.

Overall, through hedged by FPSE 100 index, the reduction of the volatility of S&P 500 index
is at 40.88% using the rolling window model with a window length of 250 days while at
38.67% using EWMA model with the decay factor of 0.94 and s12 of 0. The optimal
reduction these two models calculate is at 40.90% and 40.68% respectively, which is
achieved at a window length of 246 days and the decay factor of 0.9905. According to these
results, the hedge portfolio may be achievable and appropriate, and the validity of these two
models can be proved. In other words, the risk of investing in a long position in S&P 500
index can be reduced by hedging with a short position in FPSE 100 index. The results support
the view that the two volatility forecasting models we used can effectively derive the time-
varying hedged ratios and reduce the investment risk.
Conclusion

The results from these two models, rolling window model, and EWMA model, are similar,
which implies that it is possible to hedge a long position in S&P 500 index with a short
position in FPSE 100 index. More detailed, the hedge portfolio can decline 40.88% and 38.67%
of the volatility of S&P 500 index respectively, considering the rolling window model with a
window length of 250 days and EWMA model with a decay factor of 0.94. After
optimization, the decrease of variance can reach 40.90% and 40.68% respectively, when the
window length is 246 days and the decay factor is 0.9905. Although the result from the
rolling window model is not completely equal to that from the EWMA model, it is effective
to reduce the risk of holding a long position in S&P 500 index through hedging this index
with FPSE 100 index. In other words, holding this hedge portfolio is more reasonable and
reliable than holding S&P 500 index alone since the portfolio can significantly reduce the
volatility of investment on this index by around 40%.

However, although these two models are feasible in this study, there are several problems in
these approaches. In the rolling window model, ghost features in the estimation of volatility
exist continually and are hard to avoid. This problem is the result of that the rolling window
estimator gives observations either full weight or zero weight. It means that observation may
be ignored if it is so old or out of sample selected by window length. Therefore, there may be
an error in the estimation of volatility due to the absence of some important shock that may
impact heavily on volatility. In the EWMA model, similarly, there is an inevitable problem
that the EWMA model is not possible to represent the true process to generate volatility. It is
because that the process EWMA model established for the conditional variance is not
stationary. Specifically, this model forecasts the volatility that cannot converge to a long
value as the increase of the forecast horizon, which implies that the unconditional variance is
infinite.

Using more advanced and scientific models is an efficient method to solve these problems.
Considering the problem, full or zero weight of observation in the rolling window model, a
reasonable way is to give different weights for different period observations and make the
weight of more recent observation larger. Approaches the same as this way to weight
observation are various, including linearly declining weights, hyperbolically declining
weights, etc. The most common one is the exponentially weighted moving average which is
the second method used in this report. Similarly, the other problem mentioned before can be
solved by using an alternative model that nests the EWMA model, known as generalized
autoregressive conditional heteroscedasticity, or GARCH. Therefore, in the future research,
GARCH model might be a more effective model to estimate the volatility so that the results
from research are likely to be more accurate and correct.

Overall, although problems exist constantly in the rolling window model and EWMA model,
data collection and calculation of essential variables are corrected and successful, which leads
to the results from this research reasonable and reliable. Eventually, the effect of this hedge
portfolio can efficiently reduce the risk of holding a long position in S&P 500 index is
rational and acceptable. In the future study, a more advanced model like GARCH model that
solves problems mentioned before can be adopted to estimate a more accurate estimation.
 Reference
Baillie, R.T. and Myers, R.J., 1991. Bivariate GARCH estimation of the optimal commodity
futures hedge. Journal of Applied Econometrics, 6(2), pp.109-124.

Kumar, S.S.S., 2006. Forecasting volatility–Evidence from Indian stock and forex markets.
[online] Available at:
<http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.527.1545&rep=rep1&type=pdf>
[Accessed 28 April 2020].

Liu, C., 2016. The comparison of the forecasting models of volatility. MEc. Chongqing
University. [online] Available at: <http://cdmd.cnki.com.cn/Article/CDMD-10611-
1016730798.htm> [Accessed 28 April 2020].

McMillan, D.G., 2005. Time-varying hedge ratios for non-ferrous metals prices. Resources
Policy, 30(3), pp.186-193.

Park, T.H. and Switzer, L.N., 1995. Time-varying distributions and the optimal hedge ratios
for stock index futures. Applied Financial Economics, 5(3), pp.131-137.

Suganuma, R., 2000. Reality check for volatility models. [online] Available at:
<https://epge.fgv.br/files/1054.pdf> [Accessed 27 April 2020].

Wang, H.Q., 2015. A comparative study of volatility modeling and VaR estimation using
different models. Business Journal, (34), pp.113-115.