Module 13

Financial Risk Management

December 6, 2024

So far in this course risk has been measured by asset return’s variance. While it is
justifiable if the return is normal (or at least symmetric), it is inadequate if otherwise
(e.g., a portfolio of bonds or derivatives).

Value at Risk (VaR)

—————————
How much a portfolio  can lose over a given time horizon?

VaR: statistical estimation of such a lose (with a given confidence level  )

P (  VaR )  1  

VaR for a single asset:

To illustrate the idea and also make it analytically tractable, assume the return on an asset
S is normal with drift rate  and annualized standard deviation  . Over a period  t

 St St   t  S t
Rt ,t  t    N (  t ,  2 t )
St St

P & L  St  t  St  St Rt ,t  t

 VaR 
    t 
VaR R   t St
P ( P & L  VaR )  P ( St R  VaR )  P ( R   )  P  
St   t  t 
 
 

R   t
follows standard normal distribution N (0,1)
 t
For a short time horizon  t , ignoring the drift term (as  t is small), for a given
confidence level  ,

1
 VaR VaR
   t 
St St
P ( St  VaR )  1    N 1 (1   )   ,
 t  t

 VaR   St  t N 1 (1   )

N 1 (.) = inverse standard normal distribution function

 = 99%: N 1 (1   ) = -2.326342

 = 97.5%: N 1 (1   ) = -1.96

 = 95%: N 1 (1   ) = -1.644853

Example:
On December 5, 2024, TSLA, BA closed at $369.49 and $156.67 respectively. Their
volatility estimated using long (2 year) and short dated (6 month) historical data are 59%,
68% (TSLA) and 31%, 34% (BA) (source: Yahoo Finance).

1
For 1-day VaR we use annualization factor 250 so that t  . We compute 1-day
250
VaR:

2
 TSLA $ 369.49
vol\alpha 99% 97.50% 95%
59% 32.0745 27.0230 22.6784
68% 36.9672 31.1451 26.1378

BA $ 156.67
vol\alpha 99% 97.50% 95%
31% 7.1458 6.0204 5.0525
34% 7.8374 6.6030 5.5414

More general case:


VaR   St  t    t N 1 (1   ) 
Notice that real world, instead of risk neutral, drift is used above.

VaR for a portfolio of assets:

Consider portfolio
n
 n
wi Si n
   wi Si  R    Wi Ri , Wi  , W i 1. (see Module 11)
 n
i 1 i 1
wS
i 1
i i
i 1

Ri normal  R normal  for a short time horizon t :

VaR      t N 1 (1   )

n n
   t N (1   )
1
WW   
i 1 j 1
i j ij i j

n n
   t N 1 (1   )  w S w S   
i 1 j 1
i i j j ij i j (see Module 11)

Example:
Consider a portfolio consisting 0.5 share of TSLA and 1 share of BA. At current prices
the two stocks are weighted at similar level and its current value = 341.42 is comparable
to a single share of TSLA. Their correlation estimated from the long and short dated
historical data are 0.2 and 0.23 respectively. Using the above data for TSLA and BA we
compute 1-day VaR:

3
 corr\alpha 99% 97.50% 95%
0.2 9.4087 7.9269 6.6525
0.23 10.8363 9.1297 7.6619

VaR for a portfolio of derivatives:

Let  be a portfolio of derivatives with underlying assets {Si } , i = 1, 2, …, n. Let


i  , the delta of the entire portfolio w.r.t. Si
Si
n n
     i Si    i Si Ri
i 1 i 1

n n
 VaR    t N 1 (1   )   S  S   
i 1 j 1
i i j j ij i j

“  ” because it ignored both higher order terms in  Si (non-linearity) and changes in

other parameters such as interest rate, volatility.
 to improve VaR calculation one may add additional greeks and higher order terms
For derivative products we have realized and option implied volatility:

 Option valuation  implied volatility

 Asset price movement  forecasted realized volatility, may be different from the
option implied volatility

Monte Carlo (MC) VaR:

Widely adopted VaR methodology due to its flexibility in treating various financial assets
and products:

 Identify the variables affecting the portfolio, such as asset prices, interest rate,
volatility

 Select pricing models and determine model parameters (e.g., using lognormal model
for asset prices, a mean-reversion stochastic process for interest rate, using
econometric methods to determine their drift rates, covariance matrix, using BSM
model for vanilla options etc)

 Simulate the market movement to the desired time horizon

4
 Value the portfolio using the selected models at the generated future market scenario

 Compute the portfolio P&L by subtracting its current value

 Repeat the above process for many random paths to obtain an estimated distribution
of the P&L

 VaR can be read from this P&L distribution as the 1   quantile of the simulated
data
Example:

Simulate 1000 scenarios, order the resulting P&L. For  = 99%, take the last 10th value.
For  = 95%, take the last 50th value.

MC VaR is

 forward looking – using forecasted parameters

 model dependent – explicit assumptions on the stochastic processes of market

variables, P&L on positions of derivatives are estimated with valuation models

 computationally expensive for large portfolios

Historical VaR:

 use historical asset prices instead  assets under consideration need have adequate
historical data

 backward looking, which period of the past one should look at?

 less model dependency – no assumptions on market dynamics, no parameters to

estimate

 not completely model independent – illiquid derivatives may be “marked to model”

historically

Example:
Consider TSLA and BA daily return for the past five years (computed from close prices
during December 6, 2019 – December 5, 2024). There are 1257 daily return data points
(source: Yahoo Finance).

5
The historical 1-day VaR computed from these data:
99% 97.50% 95%
TSLA 22.8500 17.6000 14.3167
BA 14.5833 11.4100 8.4200

For the above portfolio consisting of 0.5 shares TSLA and 1 share of BA, the 1-day VaR
is
99% 97.50% 95%
VaR 22.5717 16.1983 12.8350

Expected Shortfall (ES):

Also called conditional VaR (CVaR), which address the question:
when market fall into the extreme situation indicated by VaR, how much one can lose?
1
1
ES 
1  VaR d ,
0
VaR = VaR at confidence level 

Example:

Simulate 1000 scenarios, order the resulting P&L. For  = 99%, take average of the last
10 values. For  = 95%, take average of the last 50 values.

6
Example:
Using the historical data above for TSLA and BA we compute 1-day ES:

99% 97.50% 95%

TSLA 29.5323 24.3832 20.0137
BA 25.9938 18.3295 14.0755

For the above portfolio consisting of 0.5 shares TSLA and 1 share of BA, the 1-day VaR
is

99% 97.50% 95%

ES 30.7618 23.8183 19.0239

VaR/ES are computed at trader/desk/division/region/firm level to serve as

 benchmark for business – allow comparison of risk in different lines of business

 potential loss estimate

 capital cushion

Credit Risk
—————
Credit rating:
Major US rating agencies: Moody’s, S&P and Fitch:

Moody's S&P/Fitch
Aaa AAA
Aa AA Investment
A A Grade
Baa BBB
Ba BB
B B
Caa CCC High Yield
Ca CC
C C

Finer grades between Aa/AA and Caa/CCC: Aa1, Aa2, Aa3/AA+, AA, AA-, etc.

7
Historical default probability:
Average cumulative default rates (%), 1970–2015 (Source: Moody’s):

Term (yrs): 1 2 3 4 5 7 10 15 20
Aaa 0 0.011 0.011 0.031 0.087 0.198 0.396 0.725 0.849
Aa 0.022 0.061 0.112 0.196 0.305 0.54 0.807 1.394 2.266
A 0.056 0.17 0.357 0.555 0.794 1.345 2.313 4.05 6.087
Baa 0.185 0.48 0.831 1.252 1.668 2.525 4.033 7.273 10.734
Ba 0.959 2.587 4.501 6.538 8.442 11.788 16.455 23.93 30.164
B 3.632 8.529 13.515 17.999 22.071 29.028 36.298 43.368 48.071
Caa–C 10.671 18.857 25.639 31.075 35.638 41.812 47.843 50.601 51.319

Default recovery:
Recover assets from default via

 Direct negotiation (e.g., debt modification, reorganization)

 Bankruptcy court
Amount can be recovered depends on the state of the debtor and the seniority of the debt
Terminology:

 Recovery rate (R): % of asset notional that can be recovered

 Loss given default (LGD)

A model for default:

Let the probability of survival (without default) up to time t is Ps (0, t ) . Assume that
default is a random arrival process with rate  so that the probability of default between
(t , t   t ) is  t , then the probability of surviving to t   t is

Ps (0, t   t )  Ps (0, t )(1   t )

 Ps (0, t   t )  Ps (0, t )   Ps (0, t ) t

dPs (0, t )
   dt
Ps (0, t )

 Ps (0, t )  e  t ( Ps (0, 0)  1 )

 : “hazard rate”  can be thought as default probability per unit time (default rate).

8
If  is time dependent,
t


 u du
Ps (0, t )  e 0

A simple one-period model for bonds:

Consider a zero coupon bond B with maturity T . Let R be the % can be recovered from
the bond notional N , r be the risk free rate (continuous compounding). In a single
period setting:

B (0, T )  NPs (0, T )e  rT  NR (1  Ps (0, T ))e  rT

 N  e  T  R (1  e  T )  e  rT

Example:

A zero coupon bond with notional N = 1000 maturing in 5 years. Assume the risk free
rate r = 5%, default (hazard) rate  = 8%, recovery rate R = 40%
 DF = 0.7788, Survival Probability = 0.6703
 bond price = 624.75

One may also deduce default rate  from the bond market price and recovery rate R .

If B has credit spread s so that B (0, T )  Ne  ( r  s )T

 e sT  e  T  R (1  e  T )

In the above example, s = 4.41%.

When T is small:

s

1 R

More general case:

Assume R is constant, independent from the timing of default, and the recovery is
instantaneous. Assume also that  is independent from r .
T t


 ru du T

 ru du
B (0, T )  NPs (0, T )e 0
 NR  t Ps (0, t )e 0
dt
0

9
 T t


 ( u  ru ) du T

 ( u  ru ) du
 Ne 0
 NR  t e 0
dt
0

1st term: probability of survival till maturity  PV of the notional

2nd term: sum of ( probability of surviving till t but default during (t , t  dt ) 

PV of recovery value )
Caveats:

 Recovery is not instantaneous in practice: bankruptcy proceeding could be a lengthy

process

 Amount can be recovered may be negatively correlated with likelihood of default

  and r may be (correlated) stochastic variables

How to determine  ?

Assume R can be estimated through fundamental analysis, then {B, R}  

 “reduced form” model

Firm Value (FV) model (Merton, .... )

FV = asset + liability
At maturity of liability
- asset > liability ➔ shareholders pay off or refinance debts

➔ firm's business continues beyond the maturity date

- asset < liability ➔ shareholders walk off the debt obligation

➔ bankruptcy liquidation, debt holders collect remains of the firm

 Equity = call option on FV

Firm value VF may be modeled as a stochastic variable, e.g., a lognormal process

10
dVF
  F dt   F dWt
VF

➔ Valuing call on FV in the same way as pricing call on equity

equity price, volatility ➔ FV, FV volatility

➔ value of debts, including convertible bonds

 Elegant, unified treatment across capital structure (senior, junior debts, convertible,
preferred, common equity) of the firm
 "Structural Model'

Example:
Assume the firm is capitalized with common equity S and debt B maturing at time T

VF  VS  VB

Here VS and VB represent the aggregated amount of equity and bond respectively. Let

N B be the aggregated notional of bond.

Payoff at time T:

 VB  VF VF  N B  VS  0 VF  N B
 , 
VB  N B N B  VF VS  VF  N B N B  VF

 share holders ➔ long a call on FV with strike N B

 debt holders ➔ long FV, write the call ➔ short a put on FV

Assume S pays no dividend and B is a zero coupon bond, apply BMS to VS

VS  VF N ( d1 )  N B e r (T t ) N (d 2 ) ,

VS
 SVS   FVF  N ( d1 ) FVF (Ito’s lemma applied to VS (VF , t ) )
VF

where r is risk free rate,  S is the equity volatility,

11
 V    F2 
ln  F r   (T  t )
N   2 
d1   B , d 2  d1   F T  t
F T t

 {VS ,  S }  {VF ,  F }  VB

(See Hull 24.6 (p 548-549) for a numerical example)

A simple refinement:

Assume the firm may default prior to T when VF (t )  L , where L can be either greater or
less than N B

➔ equity = KO option on FV where L may be deduced from equity option skew

(option on equity ➔ compound option on FV)

Types of credit risk:

1) credit risk inherited in security valuations such as corporate bonds  it can be treated
as parts of “market risk”
2) lending risk: amount is known, bilateral agreement but risk is one sided
3) counterparty risk: bilateral risk, amount is uncertain

Example: consider a long forward contract f with entity A on a non-dividend paying

stock S with forward price K and maturity T :

f t  St  Ke  r T t 

St  Ke  r T t   counterparty risk

St  Ke r T t   no counterparty risk (but A has counterparty risk with you)

 Amount at risk varies with St

12
Credit exposure:
Credit exposure is the amount one stands to lose in the event of counterparty default. It
depends on both time horizon and future market scenarios.
Credit exposure = max(derivative value, 0)

 Potential future exposure (PFE, for a given confidence level)

Similar to VaR, PFE of a portfolio of bilateral trades  at a given confidence level
 may be expressed as
P (  PFE )  1  

 Limit on trades, at counterparty and portfolio level

 Expected exposure (EE)

Example:

EE at time t  T for a long forward contract seen from time t0  t :

EE  E  max( St  Ke  r (T t ) , 0)   c( St0 , t0 , Ke  r (T t ) , t )e r (t t0 )

Note: here the expectation E[.] is taken w.r.t. the real world probability measure
instead of the risk neutral probability as in option pricing. Hence to utilize the
analytical results for a call option c one needs to separate the drift rate (real world
drift  ) and the discount rate r

13
Credit Valuation Adjustment (CVA):
Also called counterparty value adjustment, it is the cost of hedging default risk of the
counterparty of the transaction.
Let R be the recovery rate, EE be the expected exposure of the derivative transaction
(maturing at T ) in question, t be the counterparty default intensity, r the risk free
interest rate, trade notional N = 1,
t t
T 
 ru du T 
 ( u  ru ) du

CVA(t0 )   (1  R ) EE (t0 , t )t Ps (t0 , t )e t0

dt  (1  R )  EE (t0 , t )t e t0
dt
t0 t0

In the above it is implicitly assumed that R is constant and counterparty default is

uncorrelated with the derivative valuation and interest rate level.

Other adjustments (xVA):

 Debit value adjustment (DVA) - CVA from the counterparty's perspective

 Funding value adjustment (FVA) - adjustment due to funding cost/benefit

 Margin value adjustment (MVA) - funding costs of the initial margin

 Capital value adjustment (KVA) - cost of holding regulatory capital

Mitigating counterparty risk:

Counterparty risk is important when you have a large OTC derivative portfolio. To
mitigate counterparty risk one may resort to

 Netting

 Collateral

 Contractual clauses

 Hedging
 CVA desk

 Central counterparty (CCP)

14