Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
Module II: Key Risks and their Measurement
Section B: Market Risk
Chapter 3: Value at Risk for Trading Book
Prof. Sanjay Basu
Structure
1. Introduction
1.1 Sensitivity measures: Weaknesses
1.2 VaR basics
2. VaR Case Study
2.1 Data issues in Var estimation
3. VaR Applications
3.1 Internal estimate of capital charges
3.2 Reflect market conditions
3.3 Monitor correlations
3.4 Limit setting for market risk
3.5 Optimal RAROC
4. Weaknesses of VaR
5. Conclusion
1. Introduction
The market risk of trading positions was initially measured with different sensitivity
measures. This created serious issues for the top management at large and well-
diversified banks and FIs, which found it difficult to get a broad perspective of the
organizational risk profile. Legend has it that the former Chairman of J.P. Morgan,
Dennis Weatherstone, wanted a single number on his table, by 4.15 p.m. every day,
which summarized the risks across all his positions. At that time, J.P. Morgan had more
than 100 trading positions across the globe6. In response, his team came up with the
concept of Value-at-Risk (VaR) which quickly replaced most security-specific sensitivity
measures for the trading book.
Page 1 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
1.1 Sensitivity Measures: Weaknesses
Sensitivity measures of market risk are simple, but have many drawbacks. These are:
I. These apply arbitrary shocks to risk factors, to estimate changes in the value of
securities. For instance, modified duration assumes that yields change by 100 basis
points. The additional (non-linear) effect of larger changes in interest rates is
captured by convexity. Similarly, the market beta approach assumes that stock
market indices change by 1%, for estimating potential losses on an equity portfolio.
As discussed later, such shocks may not be related to actual movements in market
risk factors.
II. These do not take into account the correlations between financial returns in a
portfolio and the resultant diversification effects. For instance, the modified
duration-convexity method is applicable only to a parallel shift in yields. This means
that the technique works when rate shocks are the same for all yield curve
segments. In other words, correlations between yield shocks are assumed to be
perfectly positive. However, fluctuations in interest rates (e.g. as summarized in the
RBI reviews of monetary policy from time to time) are not equal across all maturity
segments. This means that interest rates for all tenors may neither move to the same
extent nor in the same direction. How do we estimate portfolio losses from such rate
shocks? Likewise, the market beta approach assumes that the equity index is the
only source of risk for all shares in the portfolio. As a result, the method works when
the correlation between all pairs of stocks in the portfolio is close to unity.
III. These do not associate a probability measure with the assumed shocks and the
resultant losses. For instance, a large interest rate shock may be used to estimate
losses on the bond portfolio. Or, a sharp decline in the stock market index may be
conceived, to estimate the hit on the equity positions. What action should the bank
take on these forecasts? How realistic are such predictions? Are these the maximum
possible losses? What is the chance that larger shocks could occur? In the absence of
more information, the bank may not be in a position to reduce exposure, revise
limits or raise additional capital, if the shocks and losses are considered to be
implausible.
IV. Lastly, sensitivity measures for individual positions cannot be added up across the
risk categories of fixed income securities, equity, forex and commodities. Thus, we
do not obtain an aggregate risk measure for the entire trading portfolio. Moreover,
we cannot estimate the extent of diversification benefits, across the various
segments in the portfolio, since the risk measures are different for different
securities. At best, we can add up loss estimates across different trading desks – we
are forced to assume that losses are perfectly correlated.
1.2: VaR Basics
Value at Risk (VaR) is a quantitative estimate of market risk that tries to overcome such
limitations of sensitivity measures. It is a single number which summarizes the
Page 2 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
downside risk (potential portfolio losses) of a bank, under normal conditions, from
movements in financial markets. It is the maximum loss over a target horizon such
that there is a low, pre-specified, probability that the actual loss will be larger3 .
For instance, if the 99% daily VaR for a bond portfolio is Rs. 1000, it means that there is
a 99% chance of losing at most Rs. 1000 over the next day. In other words, there is a 1%
chance (probability of violation α = 1%) that actual losses may exceed Rs. 1000 between
today and tomorrow. However, the VaR estimate does not tell us by how much actual
losses may exceed Rs. 1000, when and if a violation occurs. This is a serious limitation of
VaR and will be dealt with later, in detail.
The time horizon (holding period) in this example is chosen to be one day, based on the
assumption that it is possible to liquidate or hedge the portfolio without a sharp decline
in price within a day. Longer holding periods of 5 or 10-days can also be used in the
estimation of VaR and require prediction of price movements over the longer horizon.
VaR estimates increase with longer holding periods, since the cumulative losses in a
falling market, which is difficult to exit, are expected to increase with the time horizon.
The Square Root of Time rule is usually used to scale up the daily VaR to a t-day VaR. So,
if we want to compute a 10-day VaR from the daily VaR figure, when daily shocks are
identical and uncorrelated, the formula is:
10-day VaR = Daily VaR x √10
The choice of the pre-specified probability α is again left to the organization and its risk
appetite. A bank which has a low risk appetite would choose a low value of α in order to
obtain a conservative estimate of VaR. However, it may not be able to choose the highest
possible confidence level for holding capital, because of the (present and future) cost
associated with such a decision. Alternatively, if it wants to hold capital at a very low
confidence level, the regulator may specify a much higher threshold, to ensure that
depositors are adequately protected from large shocks under normal conditions. Ceteris
paribus, the lower the value of α, the higher the confidence level (1- α), the larger will be
the VaR estimate. Typical values of α are 0.1%, 1%, 2.5% and 5% and the corresponding
typical values of the confidence level for VaR estimates are 99.9%, 99%, 97.5% and 95%
respectively.
In interpreting the VaR numbers, it is crucial to keep in mind the probability α and the
holding period t. Without them, VaR figures are meaningless.
In order to compute VaR, we need to identify the basic market rates (risk factors) and
traded prices. Then, we compute returns from the price data, for three reasons: (i)
returns are comparable across positions (ii) returns are comparable over time and (iii)
returns capture the severity and direction of historical shocks. The next step involves
estimating the empirical or statistical distribution of returns, for the various segments
of the portfolio. Then, we aggregate losses across positions, using implicit or estimated
correlations among returns, to determine the potential future changes in the value of
the portfolio. VaR represents the possible portfolio (and position-wise) loss at a pre-
specified confidence level, over a given horizon.
Page 3 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
In the next section, we illustrate the superiority of VaR estimates, over sensitivity-based
loss forecasts, with an example. We also discuss some data issues for constructing VaR
models.
2. VaR: A Case Study
The following discussion is on the investment portfolio of a bank, which has used
Modified Duration-convexity to measure losses. We analyze the implications of using
VaR instead.
Table 1: Summary Risk and Return Indicators for bond portfolio of Bank A.
Coupon/ Coupon/
MD-conv (MD- (99% 10-day (VaR-
Coupon MD Conv effect conv) VaR) /MV MV)
31.3.06 8.12% 4.10 27.34 -3.82% 2.12 -2.64% 3.07
30.6.06 8.35% 4.61 31.61 -4.30% 1.94 -8.34% 1.00
30.9.06 8.39% 4.50 30.13 -4.20% 2.00 -2.79% 3.00
31.12.06 8.20% 4.42 31.07 -4.11% 2.00 -2.04% 4.01
31.3.07 8.16% 3.97 26.44 -3.70% 2.20 -2.49% 3.28
30.6.07 8.02% 3.91 24.74 -3.67% 2.19 -2.76% 2.90
30.9.07 7.76% 3.64 22.78 -3.41% 2.27 -1.62% 4.80
31.12.07 7.70% 3.62 22.68 -3.39% 2.27 -1.50% 5.14
How is VaR different from the sensitivity measures? Table 1 compares the modified
duration –convexity (MD-Conv) based approach with VaR-based analysis. It shows that
when the risk (or possible loss) is measured with modified duration-convexity, it tends
to fluctuate between 3.39% and 4.30% of portfolio value. However, when potential loss
is measured in terms of 99% 10-day VaR, it is much more volatile - between 1.5% and
8.34% of portfolio value. In what follows, we argue that a sharp difference between the
results obtained in terms of VaR and modified duration-convexity can always be
associated with a large nonparallel shift of the yield curve. For instance, between
31.3.2006 and 30.6.2006, the portfolio loss (in terms of duration-convexity) was
stronger than the previous quarter, up from 3.82% to 4.30%. The coupon–to-loss ratio
(a simple measure of risk-adjusted return) fell from 2.12 to 1.94. During the same
period, as a proportion of portfolio value, 99% 10-day VaR rose much more sharply
from 2.64% to 8.34%. As a result, risk-adjusted return fell from 3.07 to 1.00. An extract
from the RBI’s First Quarter Review of Monetary Policy in 2006-07 summarizes the
behaviour of market rates during the period:
Page 4 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
The yields in the Government securities market hardened further during the first quarter
of 2006-07………….The spread between 1-10 year yields increased to 113 basis points at
end-June 2006 (from 98 basis points at end-March 2006).
Even as portfolio maturity and duration went up, in the face of rising yields, longer
tenors got stronger rate shocks during this period. As a result, portfolio value fell more
sharply when tenor-specific discount rates (spot rates) were used and portfolio VaR
shot up. When YTM was used as the discount rate, the portfolio value fell less as
captured by the modified duration-convexity approach. The sharper rise in VaR,
compared to modified duration-convexity, not only indicates that rates had risen but
also that the yield curve had steepened. The bank should actually have reduced
portfolio modified duration by offloading longer-maturity bonds.
In contrast, the bank had an opportunity to invest more in longer-duration bonds
between 30.6.06 and 30.9.06. However, it reduced the maturity of its portfolio, the
modified duration-convexity impact fell from 4.30% to 4.2% and the portfolio coupon
rate rose marginally from 8.35% to 8.39%. As a result, risk-adjusted return rose from
1.94 to 2.00. However, since 99% VaR fell sharply from 8.34% to 2.79%, risk-adjusted
return (VaR-based) tripled from 1.00 to 3.00. An extract from the Mid-Term review in
2006-07 indicates why VaR fell so sharply:
“The yields in the Government securities market eased during the second quarter of 2006-
07, reversing the hardening trend witnessed in the first quarter………. The spread between
1-10 year yields narrowed from 113 basis points at end-June 2006 to 94 basis points at
end-September 2006”
As yields softened during the quarter and long-term rates fell more than short-term
rates, the bank could have earned a higher coupon rate at lower risk (as captured by
VaR) had it increased its portfolio maturity and duration. The benefits of softer rates,
especially at longer tenors, are reflected in a sharp fall in VaR. But the reduction in
modified duration is based on an assumed rise in yields, rather than actual rate
movements.
To sharpen the distinction between the two risk measures, let us also compare results
between 30.6.2007 and 31.12.2007. The modified duration of the bond portfolio went
down from 3.91 years to 3.64 years, between 30.6.2007 and 30.9.2007, and further to
3.62 years by 31.12.2007. The coupon rate fell from 8.02% as on 30.6.2007 to 7.76% on
30.6.2007 and further down to 7.7% on 31.12.2007. This again means that the bank was
anticipating a rise in interest rates during this period. As a result, its risk-adjusted
return (modified duration-convexity method) rose marginally from 2.19 to 2.27
between 30.6.2007 and 30.9.2007 and remained at that level even on 31.12.2007. In
contrast, as the VaR fell from 2.76% to 1.62% between 30.6.07 and 30.9.07 and further
to 1.5% on 31.12. 07, VaR-based risk-adjusted returns shot up from 2.9 to 5.14 between
end-June and end-December 2007. To resolve the large difference in results, we present
extracts from the Mid-term and Third Quarter Reviews of the RBI in 2007-08, on the
behaviour of interest rates during this period:
Page 5 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
Mid-term Review: Yields in the Government securities market softened during the second
quarter of 2007-08……….. The spread between 1-10 year yields was 52 basis points at end-
September 2007 as compared with 65 basis points at end-June 2007.
Third-Quarter Review: As on January 23, 2008, the 10-year yield was 7.42 per cent, 55
basis points lower than that at end-March 2007. The spread between 1-10 year yields was
19 basis points at end-December 2007 as compared with 52 basis points at end-September
2007.
Now it becomes clear again that the bank should have increased the maturity profile of
its bond portfolio during this period – expecting a reduction in interest rates. Moreover,
since the yield on longer-term bonds was likely to fall more (as evidenced by a steady
decline in the 1-10 year spread), it made sense to buy them for higher coupons. This is
exactly what a sharp decline in VaR for these two quarters, in Table 1, also suggests.
Since VaR captures the actual movements in interest rates, rather than assuming rate
shocks as in the duration approach, a fall in rates would tend to reduce VaR as well.
Since the opportunity gains are higher at the longer end of the maturity spectrum, VaR
would fall more when the yield curve flattens more.
2.1: Data Issues in VaR estimation – Basel II recommendations
The results of a model are as good as the data that goes into it. Since VaR is completely
based on historical data, the quality of the sample is of utmost importance. The data
should be regularly audited to ensure that the inputs are free of errors. For internal
data, the bank’s audit functions should ensure that information provided to the model
agrees with e.g. the bank’s general ledger data. Since external data is typically derived
from multiple sources, the bank has to decide from which source it will capture the data,
at what frequency and in what format. For instance, does the bank capture yield curve
data from FIMMDA, NSE or CCIL? Why does it prefer one source over others?
Data inputs can also be made to pass through simple and inexpensive automated filters.
For instance, a bank might define business rules to check the integrity of data (e.g.
interest rates should lie in a certain range during a certain interval or portfolio duration
cannot exceed 50 years). There would always be a trade-off between integrity and
completeness of data that is collected from various business units. The inputs might also
be inspected by experienced personnel, who possess an in-depth understanding of the
possible gaps in the dataset.
The revised Basel II guidelines on the Internal Models Approach to Market Risk Capital
Charges5, insist on a comprehensive dataset, which is periodically reviewed to align it
with recent market developments. The Basel II data requirements for VaR models are as
follows:
• The sample period should contain at least one year (250 days) of daily observations.
• Data sets and parameters (e.g. volatilities and correlations) should be updated at
least once every three months.
Page 6 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
• A rigorous and detailed stress testing programme must be in place.
• Daily backtests, which compare 99% VaR estimates with future profits and losses,
using the most recent twelve months of data, must be conducted for truthful
assessment and reporting of market risk capital.
3. VaR Applications
VaR has been widely applied, ever since its inception in the early 1990s, in all risk
categories at banks and financial institutions. Some of the most important uses of VaR
are:
3.1 Internal estimate of capital charges: While proposing Market Risk capital
charges for banks way back in 1995, the Basel Committee had realized that more
sophisticated institutions should be allowed to use their own data and models, to
forecast the size of losses from trading activities and create their own buffers. The
regulator would intervene only with a few multipliers and periodic model validation, to
ensure that the cushion was strong enough to withstand large losses in normal financial
markets. VaR emerged as a natural measure for such Economic (or Internal) capital
estimates, because it gave better banks the flexibility to choose their own portfolio
valuation and loss prediction models. The stakeholders immediately knew the degree of
protection on offer – if the 99% confidence was chosen, depositors would be protected
against 99% worst losses over a given horizon. As the revised Basel II (RBI 2010)
guidelines showed, it is also possible to compute VaR-based capital charges under
stressed conditions. This was in stark contrast to the standardized approach, for which
the regulator prescribed the valuation methodology, the risk estimation model and the
yield shocks for various tenors. Since the regulatory shocks are subjective, the adequacy
of capital, against large losses, might be in doubt under the standardized approach.
3.2 Reflect current market conditions: The VaR estimates are based on recent
historical data and capture fluctuations in current risk factors. Therefore, as the
preceding example shows, loss estimates portray market trends. In contrast, as
discussed, sensitivity–based loss forecasts may be based on hypothetical shocks. Hence,
VaR-based trading strategies can be more quickly adjusted, in line with prevailing
market movements, than duration or beta-based techniques. They will also be more
realistic.
3.3 Monitor correlations: Portfolio VaR depends on implicit or explicit
correlations. If such estimates are monitored over time, banks can be warned of
impending financial crises. This is more likely for those segments, which experience a
steady rise in correlations. This would be perceived as good news in rising markets – an
indicator of simultaneous profits across positions. However, this also means that losses
may occur together when markets crash. Regulators may use such trends to gauge
systemic risk as well.
Page 7 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
Recent experience from the subprime crisis confirms this point. A recent report4 found
that correlations between stock returns of banks rose steadily before the crisis, while
there was no such pattern for non-banks. The following table is borrowed from their
paper:
Table 2: US stock market correlations, 1988-20074
1993-
1988-1992 1997 1998-2002 2003-2007
Banks 0.261 0.373 0.551 0.588
Firms 0.375 0.233 0.246 0.311
When stock markets were booming, all banks gained together. However, participants
failed to realize that they were likely to collapse at the same time as well. A simple VaR
model, for the US banking sector as a whole, could have captured the systemic risk. But
no one was ready to bell the cat, in the hope that the good times would roll on forever.
3.4 Limit setting for Market Risk: If the available or planned capital buffer is VaR-
based, then reallocation of portfolio weights will ensure that aggregate losses, in future,
do not erode the buffer. Since capital is the scarcest and costliest resource, it needs to be
monitored and protected from very large losses. This is the main purpose of limit
setting – to preserve the capital at risk. We present a simple example below, to make the
point. We create a portfolio of government bonds, USD and NIFTY index and estimate
99% VaR for each security. Assuming that the capital buffer is equal to the portfolio VaR
estimate, the limit may be given by:
Table 3: VaR-based Trading Limits
99% VaR VaR Limits
Debt:
7592029 281.8941 212.2590
7962025 150.6219 113.4144
8272020 149.6978 112.7186
7682023 113.1199 85.1763
7492017 15.5406 11.7017
7952032 484.1932 364.5850
USDINR 7.3545 5.5378
NIFTY Index 1027.2086 773.4615
Portfolio
VaR 1678.8543 1678.8543
Page 8 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
The logic is very simple. If we add the 99% VaR estimates across all securities, the total
loss is more than portfolio VaR (which is the capital buffer). Therefore, the VaR limits
reduce the permissible losses and ensure that the aggregate does not exceed portfolio
VaR..
3.5 Optimal Risk-Adjusted Return on Capital (RAROC): Since capital is scarce, it
has to be allocated to its most productive uses. In other words, the goal is not only to
restrict losses to the available stock of capital, but also to ensure that each rupee of
capital is employed in the most profitable activity. This will not only ensure a high
return on capital, but also continue to meet shareholder expectations from the bank. A
simple example is given below:
Table 4: VaR-based RAROC for the Trading Book
Diversified Ex-ante RAROC
VaR-based Diversified,
Market Old New Estimated
Economic Ex-ante
Value Weights Weights Profit
Capital RAROC
Debt:
7592029 223.5665 120603.4363 19.441% 20.000% 18.145849 8.117%
7962025 118.5593 71997.2677 11.606% 20.000% 15.983392 13.481%
8272020 97.5147 98434.58316 15.868% 9.000% 4.920505 5.046%
7682023 87.4039 56574.72463 9.120% 20.000% 15.057081 17.227%
7492017 2.4237 31147.95471 5.021% 1.000% 0.114101 4.708%
7952032 381.6147 197431.207 31.826% 20.000% 19.108742 5.007%
NIFTY Index 771.4445 44157.75 7.118% 10.000% 15.670602 2.031%
Total
Portfolio 1678.8543 620346.9235 1 1 89.000271 5.301%
Debt 0.90
Other
Positions 0.10
Bank's Actual
Capital 1678.8543 89.000271 5.301%
The bank increases its exposure to 7.68% 2023 and 7.96% 2025, as well as to the NIFTY
index, to their respective limits. This ensures that the returns on its VaR-based capital
stock increase to 5.30%. Other securities, e.g. 7.95% 2032, may generate more profits
but also require much more capital. Hence, exposure to such bonds has to be reduced.
Page 9 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
4 Weaknesses of VaR
VaR is probably the most popular risk management tool, for banks and financial
institutions across the world. However, it is not without its flaws. Some of the
drawbacks were painfully exposed by the recent global financial crisis. This has
prompted regulators across the world, including the Basel Committee, to review VaR-
based Market Risk capital charges under the Internal Models Approach 1. The limitations
are as follows:
1. VaR is silent on extreme or tail losses: The VaR measure gives the risk manager an
estimate of the maximum possible loss, up to a given confidence level. But, it says
nothing on the size of losses beyond that threshold. For instance, after knowing only
that 99% daily portfolio VaR is Rs. 1000, a risk manger will not be able to forecast how
large losses are likely to be beyond Rs. 1000. This creates two problems for regulators
and top management.
First, banks might be lulled into a false sense of comfort. Low estimates of portfolio VaR,
at the confidence level prescribed by regulators, will reduce Market Risk Capital
Charges under benign conditions. Indeed, there might also be too few violations
(instances when actual losses exceed VaR estimates), vis-à-vis regulatory standards, as
long as markets remain favourable. However, a few large losses, beyond VaR, can make
these banks collapse in times of stress. If the portfolio is only vulnerable to very large
(high-severity) and rare (low-frequency) losses, the danger might not be reflected in
VaR estimates or Capital Charges, in calm markets.
The second issue is that traders and investment managers might have an incentive to
manipulate the system. Faced with VaR-based limits and capital charges, they might
choose to concentrate on securities with low historical risk (and low VaR), which are
subject to very large occasional losses. They might invest heavily in countries with
pegged exchange rates or highly rated bonds. These securities have low historical
volatility, which leads to low VaR estimates and capital charges. Their returns are also
higher than risk-free rates. But, as the history of financial crises shows, currencies can
often be rocked by a sudden and sharp depreciation and even AAA-rated bonds might
crash, due to an abrupt spike in credit and liquidity spreads. The upshot is that VaR-
based risk management systems can distort the composition of a bank’s asset portfolio
and expose it to large shocks from a few adverse events.
The problem is especially acute after a period of prolonged calmness, since even a very
large data sample might not then contain severe shocks. Without such observations, it
may not be possible for VaR estimates to forecast future stress events – the models are
just as good as the quality of input data. There may be a need for hypothetical scenarios.
2. VaR is not subadditive: This means that portfolio VaR need not always be less than
or equal to the sum of VaR estimates for all segments. There can be occasions when
portfolio VaR exceeds the sum of individual VaR estimates. Again, this creates an
incentive distortion for banks and regulators. A bank might then have the incentive to
report VaR-based market risk capital charges only for individual segments. The
Page 10 of 11
Course: Risk Management (Module II: Key Risks and their Measurement) NIBM, Pune
aggregate capital charge would be the sum of capital charges for all positions. The
regulator might feel that the bank is being conservative, in ignoring all diversification
benefits. However, such capital charges could fail to absorb portfolio losses at the bank
(Dowd 2005). The saving grace is that, though this outcome is theoretically possible, it
is likely to occur in very few real-life cases.
5. Conclusion
The more actively a bank trades in a volatile market, the more complex will be its
portfolio risk profile. VaR will be a better indicator of risk (and risk-adjusted return) in
such circumstances. It will tell the bank from which segment the risk arises, how it
affects portfolio losses and how to mitigate such outcomes. The bank will be better
prepared for the hidden risks in its portfolio, as well as risk interdependence among
products and the liquidity of the market, which duration or beta cannot capture. As a
result, its trading strategies will be more refined.
References
1. BIS (2016): Minimum Capital Requirements for Market Risk, Basel Committee on
Banking Supervision, January.
2. Dowd, K. (2005): Measuring Market Risk, 2nd Edition, John Wiley & Sons, Chichester,
England.
3. Jorion, P. (2015): Financial Risk Manager Handbook, 6th Edition, John Wiley and
Sons, New Delhi.
4. Patro, D.K., M. Qi and X. Sun (2013): A Simple Indicator of Systemic Risk, Journal of
Financial Stability, 9(1), 105 - 116.
5. RBI (2010): Prudential Guidelines on Capital Adequacy – Implementation of
Internal Models Approach to Market Risk, DBOD No. [Link].86/21.06.001(A)/2009-
10.
6. Saunders, A. and [Link] (2006): Financial Institutions Management: A Risk
Management Approach, 5th edition, McGraw-Hill, Singapore.
Page 11 of 11