Both models can be broadly categorized as statistical models
as opposed to the other classification, that is, economic
models. Unlike economic models, statistical ones do not rely
on investor behavior or economic arguments but solely
depend on statistical assumptions. The main assumption,
which is imposed on statistical models, is that the returns of
an asset are jointly normal, distributed independently and
identically through time. Despite this being a strong
assumption, it is empirically reasonable.
The constant mean return model is a relatively simple and
perhaps the simplest, yet provides great insight on the statistical behavior of the return of an asset; it is a well-known
model, which produces robust results. Warner and Brown
(1980) found that the model often produces results which are
very similar to more complicated models. It has become the
benchmark from which the more complex models are evaluated against. The model acts as a foundation from which
more complex econometric models and discussions have
been built upon. The mean average model is identical to the
measurement error model referred to in statistical literature
and states that an asset’s return is constant over time; this
constant can vary from asset to asset but remains constant for
the same asset for a given period. It is also associated with an
error term, which has a zero mean. The model’s simplicity
and robustness make it extremely attractive for researchers
who are conducting an event study, as more complex models
can be difficult to use especially when there are more variables a researcher must consider. More variables increase the
need to screen for robustness at every stage, which can be
time-consuming.
The other statistical model is the market model, which
relates the return of an asset to market returns. The possible
improvement that the market model may have over the
constant mean return model is the element, which removes
the return that is attributed to the market’s return. This
reduces the variance in the abnormal return, which allows
the model to be more sensitive to the effects on an event. An
important element, which needs to be considered with the
market model, is the R2 from the regression of the model.
The model will perform better at reducing the variance in the
abnormal return, which in turn leads to a better detection of
the event’s effects if the R2 statistic is higher. The major
issue with the market model is the fact that the model hinges
on the researcher’s ability to accurately estimate the relationship, which the asset has with the market. If the beta
which has been obtained by a researcher is not significant,
chances of the model underperforming are higher.
In this study, the mean average model is used in conjuncture with an event study to estimate the expected returns
so that the abnormal returns can be calculated. The model’s
results are then compared to assess which model was most
successful in capturing the data or was the most significant.
The literature suggests that currency markets respond to
political events and announcements. Moreover, they are
more volatile when the outcome is unforeseen, or the event
was not expected. Though there is plenty of literature that
analyzes the effects of different factors affecting currency
values, there is little literature on how currency markets
respond to uncertainty and how this can influence their
volatility.
2.3 Robustness Concerns
The robustness of a model should be a key concern when
conducting research; especially when quantitative methods
are employed. As stated earlier, this research employs the
event study methodology. Despite its popularity, the event
study method has its limitations, which may hinder the
ability to accurately measure the impact an event may have
on an asset. It is important that these issues are well identified to assess the risk involved with the study.
The first hurdle that needs to be crossed is successfully
defining the event date or the event period. As obvious as the
time of the event may seem, it is often not (Andersen et al.
2003). The main problem is not when the event occurred,
rather, when the market responded to the news. To identify
which dates were most affected the GBP value in relation to
a basket of currencies (GBPX) was visually mapped out
around the event date for this study. The actual event date
was 24 June 2016, which was when the EU referendum
result announcement was made. Through visual analysis, the
event dates of interest for this study were captured (Fig. 1.1).
The trend in the data seems apparent when visually
graphed. There is a clear steep drop on 24 June 2016 and this
drop continues on 27 of June (adjusting for non-trading
days). Hence, the selected event window was 24 to the 27
June 2016.
Another concern when applying the event study
methodology is testing the significance of aggregate results
(Cumulative Abnormal Returns, CAR). In the early days of
event studies, graphical depictions were the primary method
used to interpret the results of an event. The CAR would be
graphed by researchers to show how the market reacted to
the event. However, graphs are still an important and routine
element of representing results from an event study;
researchers are required to apply more complicated statistical
tests to prove the significance of their results. This research
will employ the use of a t-test for significance testing of the
abnormal returns.
Calculating returns has also been an issue, which has
been addressed in event study methodology, though many
researchers do not elaborate how they calculate their asset
returns. Generally, there are two methods, which can be
230
J. Janjusevic and W. Chegeni
as opposed to the other classification, that is, economic
models. Unlike economic models, statistical ones do not rely
on investor behavior or economic arguments but solely
depend on statistical assumptions. The main assumption,
which is imposed on statistical models, is that the returns of
an asset are jointly normal, distributed independently and
identically through time. Despite this being a strong
assumption, it is empirically reasonable.
The constant mean return model is a relatively simple and
perhaps the simplest, yet provides great insight on the statistical behavior of the return of an asset; it is a well-known
model, which produces robust results. Warner and Brown
(1980) found that the model often produces results which are
very similar to more complicated models. It has become the
benchmark from which the more complex models are evaluated against. The model acts as a foundation from which
more complex econometric models and discussions have
been built upon. The mean average model is identical to the
measurement error model referred to in statistical literature
and states that an asset’s return is constant over time; this
constant can vary from asset to asset but remains constant for
the same asset for a given period. It is also associated with an
error term, which has a zero mean. The model’s simplicity
and robustness make it extremely attractive for researchers
who are conducting an event study, as more complex models
can be difficult to use especially when there are more variables a researcher must consider. More variables increase the
need to screen for robustness at every stage, which can be
time-consuming.
The other statistical model is the market model, which
relates the return of an asset to market returns. The possible
improvement that the market model may have over the
constant mean return model is the element, which removes
the return that is attributed to the market’s return. This
reduces the variance in the abnormal return, which allows
the model to be more sensitive to the effects on an event. An
important element, which needs to be considered with the
market model, is the R2 from the regression of the model.
The model will perform better at reducing the variance in the
abnormal return, which in turn leads to a better detection of
the event’s effects if the R2 statistic is higher. The major
issue with the market model is the fact that the model hinges
on the researcher’s ability to accurately estimate the relationship, which the asset has with the market. If the beta
which has been obtained by a researcher is not significant,
chances of the model underperforming are higher.
In this study, the mean average model is used in conjuncture with an event study to estimate the expected returns
so that the abnormal returns can be calculated. The model’s
results are then compared to assess which model was most
successful in capturing the data or was the most significant.
The literature suggests that currency markets respond to
political events and announcements. Moreover, they are
more volatile when the outcome is unforeseen, or the event
was not expected. Though there is plenty of literature that
analyzes the effects of different factors affecting currency
values, there is little literature on how currency markets
respond to uncertainty and how this can influence their
volatility.
2.3 Robustness Concerns
The robustness of a model should be a key concern when
conducting research; especially when quantitative methods
are employed. As stated earlier, this research employs the
event study methodology. Despite its popularity, the event
study method has its limitations, which may hinder the
ability to accurately measure the impact an event may have
on an asset. It is important that these issues are well identified to assess the risk involved with the study.
The first hurdle that needs to be crossed is successfully
defining the event date or the event period. As obvious as the
time of the event may seem, it is often not (Andersen et al.
2003). The main problem is not when the event occurred,
rather, when the market responded to the news. To identify
which dates were most affected the GBP value in relation to
a basket of currencies (GBPX) was visually mapped out
around the event date for this study. The actual event date
was 24 June 2016, which was when the EU referendum
result announcement was made. Through visual analysis, the
event dates of interest for this study were captured (Fig. 1.1).
The trend in the data seems apparent when visually
graphed. There is a clear steep drop on 24 June 2016 and this
drop continues on 27 of June (adjusting for non-trading
days). Hence, the selected event window was 24 to the 27
June 2016.
Another concern when applying the event study
methodology is testing the significance of aggregate results
(Cumulative Abnormal Returns, CAR). In the early days of
event studies, graphical depictions were the primary method
used to interpret the results of an event. The CAR would be
graphed by researchers to show how the market reacted to
the event. However, graphs are still an important and routine
element of representing results from an event study;
researchers are required to apply more complicated statistical
tests to prove the significance of their results. This research
will employ the use of a t-test for significance testing of the
abnormal returns.
Calculating returns has also been an issue, which has
been addressed in event study methodology, though many
researchers do not elaborate how they calculate their asset
returns. Generally, there are two methods, which can be
230
J. Janjusevic and W. Chegeni