Professional Documents
Culture Documents
1) Business risk
1 daily operations
2 Includes strategic risk, macroecon risk
2) Financial risk
1 Interest rate movements
2 Greater awareness of Fin Risk Mgmt brought about by
o 1971 fixed exchange rate system broke down; shift to floating rate
(exchange rate risk)
o deregulation(inc. interest rate sensitivity)
o globalization(currency risk)
3) Financial institutions
1 Create markets and instruments to share and hedge risks
2 Provide risk advisory services
3 Act as a counterparty, assumes others’ risks
4) Derivative: derives value from underlying security, has a finite definite life, defined reference
rate, defined for a specific notional amount
5) Securities are non-zero-sum games and issued to raise capital (ie stocks and bonds) but
derivatives are zero sum and not raised to raise capital.
6) Leverage allows derivatives to be useful for hedging and speculation (low tranxn costs &
limited initial cash outlay) but also harder to assess downside risk (greater return variance)
7) Financial risk management
1 Detecting, Assessing & Managing financial risks
2 4 types:
o Mkt risk: declining /volatile prices lead to loss
Absolute( volatility of total returns) vs relative (relative to a
benchmark index or portfolio; measured by tracking error)
Directional (linear risk exposures)/non-directional
(nonlinear/neutral exposures to changes in economic or fin vars)
Basis risk: imperfect correlation between price of hedger & hedged
asset
Volatility risk: risk of loss from changes in actual or implied volatility
of market prices
o Liquidity: cannot liquidate at fair price leading to loss
Asset-liquidity risk: large position size forcing transxns to influence
price of securities.
Funding liquidity risk: inability to raise cash needed to rollover
debt/fulfill margin or collateral reqs of counterparties/meet capital
withdrawals
o Credit risk: counterparty default; monetary exposure value is f(n) of
P(default) and( Loss|default)
Exposure: value of loss that would be realized if credit event
occurred
o Credit event:change in counterparty’s ability to perform
previously agreed financial obligations
Recovery rate: % of assets that could be recovered from a
counterparty after a credit event occurs
Sovereign risk: result from country’s actions; country-specific
Settlement risk: failure of counterparty to deliver obligation AFTER
party has made its delivery
Presettlement risk is lower than settlement risk
o Operational risk: inadequate monitoring, management failure, defective
controls, fraud, human error. Particularly relevant to derivative trading.
Operational failure may increase market and credit risks
Model risk
People risk
Legal risk: risk of loss in value due to legal issues including lawsuits,
fines, penalties and/or damages
8) risk mgmt tools:
MCQ: In eqm, if the forward contract price of a firm’s output is greater than expected future price:
Ans (D) Since a long forward position has a negative expected return (promises to buy the asset at
F>E(price) ) we conclude that a holder of that position reduces systematic risk in absence of
arbitrage opportunities.
In frictionless mkts, risk mgmt does not add to shareholder value; but it can in mkt with tax,
transxn costs, and agency cost
Goal of risk mgmt is NOT to minimize risk, but to increase firm value.
o Unlike beta risk and output-price risk, risk of bankruptcy and financial distress (eg.
Costs from debt renegotiation, forgone value creating projects, stricter supplier
terms, lost sales, management time wasted etc) cannot be hedged.
losses in one period can be used to recover taxes paid in prior periods (tax loss carrybacks)
or to reduce tax liabilities in future periods (tax loss carryforwards) these make NO
adjustmt to the time val of money.
Risk mgmt strategies that smooth taxable income reduce total tax and increase firm value.
In WACC calculation, the contribution of debt is after-tax, hence the tax benefits of adding
debt to a firm’s capital structure.
For optimal capital structure, add debt to increase firm value by tax savings up to the point
where increase in potential costs of financial distress outweigh savings.
Risk-reduction strategies can decrease potential costs of financial distress and thereby
increase optimal debt-equity ratio, lowering funding costs (WACC).
Risk reduction benefiting a large shareholder may increase firm value: Investors are more
willing to hold large positions in firm stock when diversifiable risk is lower.
o Monitoring by large shareholders can decrease agency costs
o Large shareholders may have industry-specific expertise and advice for managers
Risk management can increase value through
o tax saving
o reduction of potential financial distress costs
o reduction of debt overhang probability.
o Reducing problem of info asymmetry
Management incentives (improved incentives increases firm value)
o Improved by risk mgmt that decreases nonmanagement-related risk factors in
incentive compensation contracts
o Incentive stock options value increase with increased volatility so cause less desire
to manage risk.
Other stakeholders can have better incentives to invest in firm-specific capital when firm can
hedge the risks inherent in such investment at lower cost than the stakeholders.
Firms have decreased value due to debt overhang when amt of debt prevents equity holders
from investing in positive NPV projects cos benefit to debt holders (must be paid first)
reduces the value created for equity holders.
Information asymmetry leads to higher capital costs (greater WACC)
o Investors must rely on mgmt estimates of quality of growth opportunities for which
funding is sought
o Unclear to investors extent to which firm performance is result of poor management
decisions or nonmanagement factors
o Hedging risks outside management control increases confidence of outside investors
that firm results reflect mgmt quality, reducing funding costs.
Capital Mkt Line shows linear relationship btw E(R) of a portfolio and its std dev when that
portfolio consists of a combination of mkt portfolio and risk-free asset.
The mkt portfolio is one of a continuum of portfolios along the efficient frontier tt is
composed only of risky assets
Efficient frontier is a convex curve drawing locus of all portfolios tt have the min std dev of
return given an expected return.
o If risk-free asset exists, CML is new efficient frontier.
CAPM assumes
o Investors are rational, risk-averse, end-of-period expected utility maximizers and all
have the same investment horizon
o Investors only consider mean and std dev of asset returns (ie quadratic utility
function; or normally-distributed returns)
o Investors can borrow/lend unlimited amts at same risk-free rate
o Investors have same expectations concerning returns. Information same.
o No taxes nor transaction costs and assets are infinitely divisible. (PERFECT mkts)
markets are (strong-form) efficient.
o CAPM claims that in eqm all investors hold a portfolio of risky assets that has the
same weights as mkt portfolio.
Co v M , i
o SML: E ( Ri ) =Rf + [ E ( R m )−R f ]
[ σM
2
]
o Market behavior is determined by beta (‘qty of risk’) and mkt risk premium ie
E(R_m)-R_f (‘price of risk’)
Treynor measure=
[ E ( R p )−R F ] is appropriate for comparing well-diversified portfolios
βp
Jensen' salpha=E ( R p )− RF + β P ( E ( R M )−R F ) is for comparing portfolios with same
[ ]
beta.
Treynor is more forward-looking than Sharpe; sharpe is considered better for measuring
historical performance
Portfolio that has higher Treynor but lower Sharpe measure than market is likely not well
diversified.
Sharpe can apply to all portfolios since it uses total risk.
σp
For well diversified portfolio, β P ≈
σM
Information ratio assesses if manager’s deviation from benchmark reaps an appropriate
E ( R p )−E ( R B )
return.e p=R P−R B ; σ e =tracking error ; info ratio=
p
[ σep
] =α P /σ e p
E ( R p )−R min❑ 1
Sortino Ratio = , MS D min❑= ¿
√ MS Dmin❑ N
Limitations of CAPM
o Mkt portfolio shd be portfolio with highest sharpe ratio of all, and must include all
investible assets. This assumption may hold for a globally constructed portfolio but
cannot hold when restricted to a single equity market.
o Expected excess returns for mkt are assumed to be known, based on assumptions
that (1) investors have access to same info and have homog expectations (2)
distribution probability beliefs of investors match true returns distr.
APT: multifactor regression model; “mkt” portfolio will have lowest risk of all possible
portfolios given identical factor exposures
A qualified model utilizes a mkt portfolio that is fully diversified so tt the model explains a
majority of the variation in excess returns
Finding a qualified model is easy part of implementing APT< but Factor Forecasting is hard.
Fundamental law of active mgmt: investor’s info ratio is a f(n) of dept of knowledge about
individual securities (information coefficient) and the no. of investment decisions (investor’s
breadth)
If an analyst could independently forecast 1000 stocks or 10 factors, he would need to be
10x as good at forecasting factors to produce the same Info Ratio.
Backcasting- using historical average factor returns as a forecast
Structural Model---assume an underlying factor-return relationship possible to forecast
factors using statistical methods of forecasting and verify that these forecasts are
reasonable. Not possible to do so in statistical model.
o Structural Model –3 types: Factor exposures known and factor returns forecasted,
factor returns known and exposures forecasted, both forecasted
o Statistical Model – PCA/MLE factor analysis/ Asymptotic Principal Components.
(asmpy. PC Reduces size of covariance matrix (TxT) )
o Practitioners mainly use structural models. Academics are primary users of stat
models cos they don’t want to introduce bias.
7.
Methodology
1. Formulate hypothesis
2. Collect data
3. Specify model/math representation of theory
4. Specify econometric model (Not “deterministic” , there is an error term)
5. Estimate params
6. Test model specification (i.e. choice of input vars)
7. Test hypothesis
8. Use model for forecasting
Pooled data set contains a set of cross-sxnal observations over several time periods.
Panel data: type of pooled data where researcher looks at a unit of observations such as a fmily
and records changes for each member of the family over time.
Dummy vars=categorical variables
Chebyshev’s inequality: for any set of sample/population data, regardless of shape of distrbutn,
percentage of observations that lie within k std devs of the mean is > (1-1/k^2) for all k>1
σ x std dev ofX
Relative dispersion: measured with coefficient of variation= = In an investment
X́ av . val of X
setting.CV is used to measure risk perunit of expected return
A distr that has a greater % of small deviations from the mean and a greater % of extremely large
deviations from the mean will be leptokurtic (+ve kurtosis)
3
( X i− X́ )
∗1
Skewness: n
s3
Sk =∑ ; s is sample standard deviation
i=1 n
4
( X i− X́ )
∗1
n
s4
Kurtosis=∑ ; s is sample standard deviation
i=1 n
Excess kurtosis=Kurtosis-3; positive excess kurt=leptokurtic (fat tails and more peaked) ; negative
(platykurtic) (thin tails, less peaked); Excess kurtosis>1 is large
Sampling error is the difference between a sample statistic and its corresponding popn
param.
Std error of the sample mean is the stdev of the distribution of the sample means and is
calculated as
σ s
σ X́ = ; σ ,the popn stdev is known;∨s X́ = ; s is sample stdev∧the popn stdev is unknown
√n √n
t-dist (shorter and fatter tails than normal; as n ∞, gets taller with thinner tails and
approaches normal) ; defined by a single parameter, the degrees of freedom = n-1
t dist is appropriate for constructing confidence intervals based on
1. small samples from popns with unknown variance and approximately normal distrib.
2. Unknown popn variance, large enuf n tt CLT assures sampling distrib approx normal
(similar categorization for hypothesis testing; But if n is very large, z-stat can
also be used even if popn variance is unknown)
Z 0.05=1.645 ; z 0.025=1.96 ; z 0.005=2.575 ; P ( within one std dev ¿mean )=0.68
( n−1 ) s2
χ 2n−1= 2
where s2 is sample var ; σ 20 ishypothesized value for popn varianc e is
σ0
asymmetrical and approaches normal as dof increases; used for hypothesis tests concerning
variance of a normally distributed popn.
σ 21
F(n −1 ,n −1)= 2 ; by convention σ 21> σ 22 ; generally ∈testing σ 2 is represented by estimate s 2 ; n1∧n2 are respecti
1 2
σ2
tests equality of variances of 2 popns.
F dist is right-skewed and truncated at zero on the left side. Shape determined by 2 separate
dofs. Rejection region is the right-side tail of distribution (becos of F stat computation
convention) .
F approaches normal as n increases
Square of t-stat with (n-1) dfs is F-distributed with (n-1) numerator and denominator df (?? I
think it is F_(n-1,1)
Example Problem:
A random sample of analyst earnings estimates has a mean of 2.84 and a std dev of 0.40. What can
we say about the 90% CI for earnings next period if (1) sample size, n =20? (2) n=40?
What probabilistic statement could we make at the 9)% Conf Lvl if (3) n=15, (4) n=60?
(1)& (2) : no statement cos we have NO DIRECT INFO about distribution of POSSIBLE EARNINGS of
the next period, only have estimates’ data.
(4)t-stat for a 90% CI with 59 dofs is approximately 1.671 (t_60). 90% CI is 2.84+-
1.671(0.4/sqrt(60)) . and we are 90% confident that the true mean of the population of ANALYST
ESTIMATES is within this range.
o Peaks-over-threshold (POT)
More modern EVT model than block maxima
Utilizes a distribution of extreme realizations above a high (or below a low)
threshold
Distribution of these extreme values follows the generalized Pareto
distribution (GPD)
To calculate GPD, one must establish a threshold u and calculate the
conditional probability of loss above u.
Shape parameter ξ indicates fatness of tail, along with beta, a scaling
parameter
Values of ξ>0 indicates fat tail and is assumed for GPD:
−1
ξx
(
F ( x )=1− 1+
β ) ξ
GPD exhibits a curve that dips below the normal distribution prior to the tail.
It then moves above the normal distribution until it reaches the extreme tail.
The GPD then provides a linear approximation of the tail, which more closely
matches empirical data than normal distrib.
With knowledge of distributional pptys of the tail, VaR can be computed at
high CIs regardless of distribution describing the data.
If sample of X and Y variables is a random sample, difference between the sample coefficients
and the population coefficients of regression will be random too.
Linear regression model is linear in both variables and parameters
b 1=¿
MLR assumptions:
Multicollinearity causes greater probability that we will incorrectly conclude that a variable is
not statistically significant (ie Type II error)
Sample population variance , σ^2=¿
Coefficient of multiple correlation=sqrt(R^2)
F=(ESS/df)/(RSS/df) ; the first df=k,denominator df=n-k-1, where k is no. of indep vars
Adjusted R^2 = 1-(1-R^2) x (n-1)/(n-k-1); is always smaller than R^2
Adjusted R^2 can be negative; and adding a new indep variable may either increase or decrease
adjusted R2 though it Always increases R2
Intro to VaR
VaR methods
Linear methods: replace portfolio positions with linear exposures on the appropriate risk factor
Delta-normal method is appropriate for large portfolios without significant option-like exposures.
This method is fast and efficient.
o Problems: When a distribution has fat tails (equity returns frequently exhibit leptokurtosis) ,
var will tend to underestimate the loss and its assoc. probability. Also, Historical calculation
if portfolio composition changes/unusual events in estimate period/changed econ.
Conditions inaccurate
o Results in a higher proportion of distributions with fat tails, cos of either unidentified time
variation in risk or unidentified risk factors and/or correlations
o Nonlinear relationships of option-like positions are not adequately described. Instability of
option deltas is not captured so VaR is misstated.
o If true expected return is used, VaR for diff lengths periods must be calculated independently
o Usually, calculate VaR over a short time period (Basel Accord recommends 2 week ie 10 days)
and then multiplied by sqrt(multiplying factor) to get VaR over longer time period
Full valuation methods: fully reprice the portfolio for each scenario encountered over a historical
period, or over great no of hypothetical situations developed thru historical simulation or Monte
Carlo simulation
1) Monte Carlo simulation revalues a portfolio for a large no. of risk factor values randomly selected
from a NORMAL distribution.
Advantages: most powerful model; can account for both linear and nonlinear risks, can include time
variation in risk and correlations by aging positions over chosen horizons, can incorporate additional
risk factors, nearly unlimited nos of scenarios can produce well-described distributions
2) Historical simulation revalues a portfolio using actual values for risk factors taken from historical
data.
Can use percentile method, or some form of averaging; the percentile method gives a more
conservative estimate.
Advantages are that horizon can be selected, not exposed to model risk, based on actual
price data, includes all correlation sas embedded in mkt price changes
Disadvantages are: only one event path is used which includes changes in correlation sand
volatilities that may only have occurred in that historical period; time variation of risk in past may
not reflect future; model may not recognize changes in volatility and correlations from structural
changes; slow to adapt to new events though EWMA can help mitigate; small no. of actual
observation smay lead to insufficient defined distribution tails
These full valuation approaches provide the most accurate measurements because they include all
nonlinear relationships and other potential correlations that may not be included in the linear
valuation models.
Full valuation methods may be the only appropriate methods for large portfolios with 1) substantial
option-like exposure (ie nonlinear) 2) a wider range of risk factors, or a 3) longer-term horizon.