Scenario analysis

When banks first started stress testing it was often referred to as Scenario Analysis.
This seeks to investigate the effect, i.e. the change in value of a portfolio, of a
particular event in the financial markets. Scenarios were typically taken from past,
or potential future, economic or natural phenomena, such as a war in the Middle
East. This may have a dramatic impact on many financial markets:
Ω Oil price up 50%, which may cause
Ω Drop in the US dollar of 20%, which in turn leads to
Ω A rise in US interest rates of 1%
These primary market changes would have significant knock-on effects to most of
the world’s financial markets. Other political phenomena that could be investigated
include the unexpected death of a head of state, a sudden collapse of a government
or a crisis in the Euro exchange rate. In all cases it is the unexpected or sudden
nature of the news that causes an extreme price move. Natural disasters can also
cause extreme price moves, for example the Japanese earthquake in 1995. Financial
markets very quickly take account of news and rumour. A failed harvest is unlikely
to cause a sudden price shock, as there is likely to be plenty of prior warning, unless
the final figures are much worse than the markets were expecting. However, a wellreported
harvest failure could cause the financial markets to substantially revalue
the commodity, thereby causing a sustained price increase. A strong directional
trend in a market can have an equally devastating effect on portfolio value and
should be included in scenario analyses.
Stress testing with historical simulation
Another way of scenario testing is to recreate actual past events and investigate their
impact on today’s portfolio. The historical simulation method of calculating VaR
lends itself particularly well to this type of stress testing as historical simulation
uses actual asset price histories for the calculation of VaR.
Scenario testing with historical simulation simply involves identifying past days
on which the price changes would have created a large change in the value of today’s
portfolio. Note that a large change in the portfolio value is sought, rather than large
price changes in individual assets. Taking a simple portfolio as an example: $3
million Sterling, $2 million gold and $1 million Rand, Figure 8.6 below shows 100
days of value changes for this portfolio.

It is easy to see the day on which the worst loss would have occurred. The size of
the loss is in itself an interesting result: $135000, compared to the VaR for the
portfolio of $43000. Historical simulation allows easy identification of exactly what
price changes caused this extreme loss.

This is useful information, a bank would be able to discuss these results in the
context of its business strategy, or intended market positioning. It may also suggest
that further analysis of large price changes in gold are indicated to see whether they
have a higher correlation with large price changes in sterling and rand. Identifying
the number of extreme price moves, for any given asset, is straightforward. Historical
simulation enables you to go a step further, and identify which assets typically move
together in times of market stress. Table 8.2 shows the price changes that caused
the biggest ten losses in five years of price history for the example portfolio above.
It can be seen from Table 8.2 that the two currencies in the portfolio seem to move
together during market shocks. This is suggesting that in times of market stress the
currencies have a higher correlation than they do usually. If this was shown to be
the case with a larger number of significant portfolio value changes then it is
extremely important information and should be used when constructing stress tests
for a portfolio and the corresponding risk limits for the portfolio.
In fact, for the example portfolio, when all changes in portfolio value of more than
1% were examined it was found that approximately 60% of them arose as a result of
large price moves (greater than 0.5%) in both of the currencies. This is particularly interesting as the correlation between rand and sterling over the 5-year period
examined is very close to zero. In times of market stress the correlation between
sterling and rand increases significantly, to above 0.5. 241