This preview shows page 1. Sign up to view the full content.
Unformatted text preview: E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 1 of 16 Lecture 9: Patterns of Volatility Change Copyright Emanuel Derman 2008 • 4/18/08 Regimes of volatility: how implied index volatilities move • The best average hedge? smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 2 of 16 9.1 Patterns of Volatility Change Copyright Emanuel Derman 2008 How do implied volatilities of options with definite strikes actually behave as
stock price changes and time passes? 4/18/08 The implied volatility surface of index options  a negative skew A negative correlation between atthemoney volatility and index level smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 3 of 16 Copyright Emanuel Derman 2008 But you cannot buy atthemoney implied volatility, so let’s look at the volatility of definite strikes, which is much more complex. What is going on? Here’s
some old data during periods of crisis. 4/18/08 smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 4 of 16 Copyright Emanuel Derman 2008 The S&P 500 implied volatility skew at any instant when the index is at level
S0 is approximately described by the linear approximation 4/18/08 Σ ( K, t 0 ) = Σ atm ( S 0, t 0 ) – b ( t 0 ) ( K – S 0 ) , which is not too bad an approximation when the strike is near the money.
In what follows, we will write Σ atm ( S 0, t 0 ) ≡ Σ 0 and b ( t 0 ) ≡ b , so that
Σ ( K, t 0 ) = Σ 0 – b ( K – S 0 ) Eq.9.1 This simple linear dependence on strike K tells us nothing about how implied
volatilities will vary when the index level moves away from S0; the (K – S0)
term is intended to describe only the variation in K.
Nevertheless, anyone interested in option value wants to know what will happen to the option’s implied volatility when the index moves to a level S. What
is the Sdependence of Σ ( S, t, K, T ) given its observed Kdependence? This
question is important for obtaining appropriate hedge ratios as well as options
values. You can take a look at the paper Regimes of Volatility, Risk 1999, or on
my website, to see a study of how implied volatilities actually moved in the
late 1990s during periods of market crisis. 9.1.1 Rules, Models, Theories for the Variation of Implied Volatility
Given the current index implied volatility skew, one wants to know how it will
vary with index level in the future. It’s often easier to formulate models of
change by stating what doesn’t change. In physics or mathematics, such quantities are called invariants. In the world of options trading, it’s become customary for traders to refer to what doesn’t change as “sticky.” There are at least
three different views on which aspects of the current skew are sticky as the
index moves over short times. These are models or rather descriptions of the
BlackScholes quoting parameter called implied volatility Σ rather than the
underlyer S and its stochastic evolution. Obviously, being descriptions, none of
them should be regarded as absolutely correct, especially over a long period.
Our aim will to be see to what extent these rules hold sway over the options
market at different times.
The three rules are:
1. The Sticky Strike Rule.
2. The Sticky Delta Rule.
3. The Sticky Implied Tree Rule. smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 5 of 16 Copyright Emanuel Derman 2008 9.1.2 The Sticky Strike Rule 4/18/08 Given the current skew, one possibility is that as the index moves, each option
of a definite strike maintains its initial implied volatility – hence the “sticky
strike” appellation. Obviously, this cannot hold indefinitely. This is the simplest “model” of implied volatility.
Mathematically, the sticky strike rule is Σ ( S, K, t ) = Σ 0 – b ( K – S 0 ) Sticky Strike Rule Eq.9.2 This is equivalent to assuming that Equation 9.2 holds true for any index level
S, that is, that implied volatility has no dependence on iS. The value S0 is
present simply to indicate the current skew’s dependence on moneyness.
Because of our assumption of stickiness, we have assumed that b ( t ) = b ,
independent of t. Of course, b can changes, even dramatically during crisis
periods.
Intuitively, “sticky strike” is a poor man’s attempt to preserve the BlackScholes model. It allows each option an independent existence, and doesn’t
worry about whether the collective options’ market view of the index is consistent. It models the current skew by attributing to each option of a definite strike
its own future BlackScholesstyle tree of constant instantaneous volatility.
Then, as the index moves, each option keeps the exactly the same constant
future instantaneous volatility in its future BlackScholes valuation tree by
moving the previously current tree so that its root now sits at the current index
level.
Equation 9.2 shows that the implied volatility for an option of strike K is independent of index level S, and therefore the delta of the option is the same as the
BlackScholes delta.
Table 1 summarizes the behavior of volatilities under the stickystrike rule.
TABLE 1. Volatility behavior using the stickystrike rule.l
Quantity Behavior Fixedstrike volatility: is independent of index level Atthemoney volatility Σ atm ( S ) : Σ atm ( S, t ) = Σ 0 – b ( S – S 0 )
which decreases as index level increases Exposure Δ: = ΔBS smilelecture9.fm Copyright Emanuel Derman 2008 E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change 4/18/08 Page 6 of 16 You can think of this model as representing Irrational Exuberance. When the
index market rises implied volatility falls, so that you trade atthemoney
options at progressively lower volatilities as the index rises. If you are a market
maker, and you lower the implied volatility as the market rises, this implies
that you think that the higher the market gets, the less likely a future catastrophe will occur. That’s irrational and it can’t go on for too long or you will end
up at zero volatility. 9.1.3 The Sticky Delta/Moneyness Rule
The stickydelta rule is a more subtle view of what quantity remains invariant
as the index moves. It’s easier to start by explaining the related concept of
sticky moneyness.
Sticky moneyness means that an option’s volatility depends only on its moneyness K ⁄ S ; the option’s dependence on index level and strike derives entirely
from its dependence on the moneyness. In mathematical terms, this means we
generalize Equation 9.1 to
Σ ( S, K, t ) = Σ 0 – b ( K ⁄ S – 1 ) S 0 Sticky Moneyness Rule Eq.9.3 where So is the initial index level at which the skew is first observed. Expanding this for S close to S0 close to K, we obtain to leading order in ( K – S ) for S
and K both close to S 0 :, the approximation
Σ ( S, K, t ) = Σ 0 – b ( K – S ) Eq.9.4 The sticky moneyness rule is an attempt to shift the skew as the stock price
moves by adjusting for moneyness. It quantifies the intuition that the current
level of atthemoney volatility, the volatility of the most liquid option, should
stay constant as the index moves. Similarly, for example, the option that is 10%
out of the money after the index move should have the same implied volatility
as the correspondingly outofthemoney option before the index move.
This model assumes that the market mean reverts to a definite atthemoney
volatility Σ 0 independent of market level. It’s a model of common sense and
moderation.
For a roughly linear skew, Equation 9.4 indicates that
Σ ≈ Σ(S – K) smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 7 of 16 Copyright Emanuel Derman 2008 so that implied volatility is the same for all values of S – K , and in particular of
r S – K = 0 , i.e. at the money. You can see from this, in the graph below, that
implied volatility must rise when S rises. 4/18/08 Σ ( S, K )
Σ ( 100, 80 )
Σ ( 100, 100 )
Σ ( 80, 80 ) S = 100
Σ ( 80, 100 ) 80 100 S = 80
K In the BlackScholes model, the exposure ΔBS depends on K and S through the
moneyness K/S, so that “sticky moneyness” is equivalent to “sticky delta,”
with an atthemoney option corresponding approximately to ΔBS = 0.5.
Options market participants often find it very convenient to think of the value
of the (ΔBS  0.5) as a measure of an option’s outofthemoneyness expressed
in units of its volatility until expiration. But, strictly speaking, sticky delta
means that the implied volatility must be purely a function of Δ BS , not just
ln S ⁄ K
K ⁄ S . Δ BS depends on  .
Στ
Equation 8.12 as well as the illustration above shows that in a negatively
skewed market the implied volatility for an option of strike K increases with
index level S, and therefore the delta of the option is greater than the BlackScholes delta for an option with the same BlackScholes volatility. Table 2
summarizes the behavior of volatilities under the stickydelta rule. smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 8 of 16 TABLE 2. Index Volatility behavior using the stickydelta/moneyness rule. Copyright Emanuel Derman 2008 Quantity Behavior Fixedstrike volatility: increases as index level increases Atthemoney volatility: is independent of index level Exposure Δ: > ΔBS 9.1.4 The Sticky Implied Tree Model: One Index, One Local
Volatility Function, One Tree.
You can interpret all current index options prices as determining a single consistent unique tree – the implied tree – of future instantaneous index volatilities
consistent with the current market and its expectations of future volatilities.
This consistency contrasts with the two previous stickiness rules, where each
option is described by a different BlackScholes tree for the index, because the
BS volatility changes with strike and market level, even though all options
have the same index underlyer.
Figure 9.1 shows a schematic view of the implied tree and its local volatilities
consistent with a particular implied volatility surface. These local volatilities,
which vary with future index level and future time, bear the same relation to
current implied volatilities as forward rates bear to current bond yields. 4/18/08 index level FIGURE 9.1. The implied tree corresponding to a given implied volatility surface. variable local
volatility σ(S,t)
in the future time smilelecture9.fm Copyright Emanuel Derman 2008 E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change 4/18/08 Page 9 of 16 In the tree below, local volatility increases (twice as fast as the skew, we’ve
seen) as the index level decreases, because the implied tree model attributes the
implied volatility skew to the market’s expectation of higher realized (local)
volatilities and higher implied volatilities in the event that the index moves
down. You can also approximately think of this aversion to increased volatilities on a downward index move as an aversion to downward index jumps.
Once you have determined the future index tree implied by the current skew
and current index level, you can isolate future subtrees within it at different
future index levels to compute the options market’s expectation of future
skews. This is similar to rolling along the curve of forward rates to compute the
bond market’s expectation of future yields.
Extracting Local Volatilities From Implied Volatilities.
The implied tree model allows the detailed numerical extraction of future local
and implied volatilities from current implied volatilities. However, just as the
yield to maturity of a bond is the linear average of forward rates until the
bond’s maturity, so you can think of an option’s BlackScholes implied volatility as the approximately linear average of local volatilities between the current
index level and the option’s strike. Table 3 displays a simple example in which
we use this approximation to extract future local and implied volatilities from
the current skew. For strikes not far from the money, this method works surprisingly well when compared with more exact numerical methods.
Table 3: Extracting local volatilities from a sample of hypothetical implied
volatilities
Index = 100
Strike BS Vol.
(%) Index Local Vol.
(%) 100 20% 100 20% 99 21% 99 22% 98 22% 98 24% 97 23% 97 26% The first two columns in Table 3 show the current implied volatility skew
when the index is at a level of 100. The skew is taken to be linear and negative,
increasing at one volatility point per strike point. When the index is at 100, the
100strike atthemoney volatility in column 2 is 20% per year. This is also the
value of the local volatility at index level of 100 in column 4, because local smilelecture9.fm Copyright Emanuel Derman 2008 E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change 4/18/08 Page 10 of 16 volatility at some index level is effectively the shortterm atthemoney
implied volatility at that index level. The 99strike volatility in column 2 is
21%. Therefore, the expected atthemoney (local) volatility at an index level
of 99 must be 22%; it is the average of a local volatility of 20% at an index
level of 100, and 22% at an index level of 99, that averages to 21% for the 99strike option when the index is at 100.
Column 4 shows the local volatilities corresponding to this skew, computed
using this averaging procedure. The averaging necessitates that local volatilities must increase twice as fast with index level as the implied volatilities
increased with strike level. In the bond world, the analogous statement is that
forward rates increase twice as fast with future time as bond yields increase
with maturity.
Given the local volatilities in column 4, we can use them to reconstruct the
implied volatilities at a different index level, say 99. The implied volatility is
22% at a strike level of 99, equal to the local volatility in column 4 at an index
level of 99. The implied volatility is 23% for a strike of 98, the average of the
local volatility of 22% at 99 and 24% at 98 in column 4.
Implied Volatility In The Sticky Implied Tree Model.
Table 3 illustrates the use of the one consistent implied tree. As the index level
within the tree rises, you can see that the local volatilities decline, monotonically and (roughly) linearly, in order to match the linear strike dependence of
the negative skew. In the sticky implied tree model, the implied volatility is an
average of the local volatilities between S and K, so that, in the linear approximation to the skew,
Σ ( S, K, t ) = Σ 0 – b ( K + S – 2 S 0 ) Sticky Implied Tree Model Atthemoney volatility is given by
Σ ( S, S, t ) = Σ 0 – 2 b ( S – S 0 )
This equation shows that implied volatilities decrease as K or S increases. Atthemoney implied volatility, for which K equals S, decreases twice as fast with
index level as the skew slope with respect to strike. Because volatility
decreases as you move to higher index levels in the tree, an option’s exposure
delta in the model is smaller than the BlackScholes delta of an option with the
same volatility.
In the linear approximation for local volatility models you can write
Σ ≈ f ( K + S ) with Σ a function of ( K + S ) , which tells you how to relate the
skew at different strikes and spot levels to each other via an invariance principle. smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 11 of 16 Let’s draw this for a roughly linear negative skew in the following figure. Copyright Emanuel Derman 2008 Σ ( 80, 80 ) 4/18/08 Σ ( S, K )
Σ ( 80, 100 )
Σ ( 100, 80 ) Σ ( 100, 100 ) 80 S = 80
S = 100 100 K In order to satisfy Σ ( 100, 80 ) = Σ ( 80, 100 ) with a negative skew, you can
see that implied volatility must rise as the stock price decreases.
Table 4 summarizes the variation of implied volatility in the sticky implied tree
model. Volatilities of options are anticorrelated with the index, rising as the
index falls and falling as it rises. Atthemoney volatilities respond to index
moves at twice that rate.
TABLE 4. Equity index volatility behavior in the sticky implied tree model.
Quantity Behavior Fixedstrike volatility: decreases as index level increases Atthemoney volatility: decreases twice as rapidly as index level
increases Exposure Δ: < ΔBS In this regime the options market experiences fear. The sticky implied tree
model assumes the skew arises from a fear of higher market volatility in the
event of a fall, or perhaps jumps, and assumes that after the fall, market volatility will rise twice as fast. smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 12 of 16 9.1.5 Summary of the Rules Copyright Emanuel Derman 2008 Assume the current stock price is S 0 , and that the current skew at this instant Sticky
Strike of time is linear, of the form Σ ( S 0, K ) = Σ 0 – b ( K – S 0 ) . Here is a summary
of the different “stickiness” and the models which correspond to them. General functional form
for future implied volatility Linear approximation:
Future skew when stock price is S Model with
this property
BlackScholesa Σ ( S, K ) = f ( K ) Σ ( S, K ) = Σ 0 – b ( K – S 0 ) Moneyness Σ ( S, K ) = f ( K ⁄ S ) Σ ( S, K ) ≈ Σ 0 – b ( K – S ) Stochastic volatilityb, jump
diffusion Implied tree/
local volatility Σ ( S, K ) = f ( K, S ) Σ ( S, K ) ≈ Σ 0 – b ( K + S – 2 S 0 ) Local
volatilityc Delta Σ ( S, K ) = f ( Δ ) because Σ is approximately the average
of the local volatilities between spot and strike. Σ ( S, K ) ≈ Σ 0 – b [ 0.5 – Δ call ( S, K, t, T ) ] ? or ln K ⁄ S
Σ ( S, K ) ≈ Σ 0 – b' Στ
Note that Δ is itself a function of Σ ! a. The BlackScholes model corresponds roughly to the sticky strike rule of thumb, but it cannot honestly accommodate a skew, because all implied
volatilities are the same irrespective of strike in the BlackScholes model. So, although people use it, it’s not really consistent from a theoretical
point of view b. In stochastic volatility models, there is another stochastic variable, the volatility itself, and so Σ ( S, K ) = f ( K ⁄ S ) only if the other stochastic variable doesn’t change.
c. Crepey, Quantitative Finance 4 (Oct. 2004) 559579, argues that the local volatility hedging is the best for equities markets, in that it gets
things right when the market moves a lot and isn’t very wrong otherwise. See next page. 4/18/08 9.1.6 Regimes of Volatility
None of these rules describe the behavior of implied volatility for long periods.
They may however be useful in describing the behavior of implied volatilities
over shorter periods. You can see more about these rules and the extent to
which the options market empirically satisfied different rules at different times
in t he paper Regimes of Volatility in Risk, 1999.
What seemed to happen is that during calm upwardtrending periods, the market satisfied the sticky strike rule, and during fearful periods it comes closer to
satisfying the sticky implied tree rule. Over time, the combination of these
behaviors leads to an atthemoney volatility that seems to revert to the same
level. You can notice this in the earlier graph of S&P implied volatilities for smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 13 of 16 individual strikes at the start of the section “Patterns of Volatility Change” on
page 2. Copyright Emanuel Derman 2008 Crepey1 has carried out a similar analysis of the FTSE 100. Note that though at themoney volatility is strongly negatively correlated with the FTSE, fixedstrike volatility remained more or less constant as the FTSE rose, and then
increased sharply as the FTSE dropped. It is difficult to generalize since volatility is fundamentally stochastic. 1. S. Crepey, Quantitative Finance 4 (Oct. 2004) 559579: 4/18/08 smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 14 of 16 Copyright Emanuel Derman 2008 9.2 Problems and Benefits of Local Volatility
Models 4/18/08 9.2.1 Inadequacy of the ShortTerm Skew
Setting aside questions of implementation, the main problem with local volatility models (for index options) is that, once calibrated to the current smile,
(which flattens in moneyness at large expirations), future shortterm local volatilities have less skew than current shortterm implied volatilities.Therefore the
shortterm future skew in a local volatility model is too flat when compared
with the index’s perennially steep shortterm skew. This suggests that local
volatility models may not be the right model for the shortterm index skew.
On the other hand, all financial models need recalibration; not one of them can
model the future accurately given an initial state of the world; that’s why, even
in BlackScholes, the implied volatility changes from day to day. Local volatility models are still popular when recalibrated regularly; they allow the valuation of exotic options consistent with the volatility surface for vanilla options,
and are widely used as a means of valuing exotics. 9.2.2 Better Hedge Ratios During Volatility Regimes
Local volatility models may well produce better hedge ratios than Black
Scholes. An argument of Crepey’s, summarized below, suggests that when the
models are regularly recalibrated, the hedge ratios produced by local volatility
models may be better than those of BlackScholes.
First, given that no model is perfect, the best hedge is the one that minimizes
the variance of the P&L of the hedged portfolio. If the replication were exact,
the variance would vanish.
Now consider a call on the index S, with value C ( S, t, K, T, Σ ( S, t, K, T ) ) , and
consider the hedged portfolio Π = C – Δ S , long the call and short Δ shares of
the index. We can now compare the effect of using local volatility hedge ratios
vs. BlackScholes implied volatility hedge ratios. The relevant hedged portfolios in the two cases are
Π BS = C – Δ BS S
Π loc = C – Δ loc S Eq.9.5 Therefore, the realized difference between the BlackScholeshedged P&L and
the localvolatilityhedged P&L is entirely due to the difference in the hedge
ratios used, so that for a stock move δ S , smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 15 of 16 δ Π loc – δΠ BS = ( – Δ loc + Δ BS )δ S ≡ ε × δ S Eq.9.6 Copyright Emanuel Derman 2008 since the change in the market value of the option is the same in either case.
In a local volatility model, in a negatively skewed market,
Δ loc = Δ BS + ∂C ∂Σ
∂C ∂Σ
≈ Δ BS +
≤ Δ BS
∂Σ∂S
∂Σ∂K and so ε = ( Δ BS – Δ loc ) ≥ 0 .
From the theory of hedging, the formulae for the respective changes in the
P&Ls during time δ t in each model, as we worked out in Lecture 2 using a
Taylor expansion, is
1
2
2
2
δΠ BS =  Γ BS S [ σ R – Σ ]δ t
2
δΠ loc Eq.9.7 1
2
2
2
=  Γ loc S [ σ R – σ ( S, δ t ) ]δ t
2 since the implied volatility of the index at level S over the next instant is
exactly the local volatility σ ( S, δ t ) in the tree at index level S for very short
expirations. (The p.d.e. in local volatility models is the BlackScholes equation
with Σ in the equation replaced by σ ( S, δ t ) .)
The BS P&L is positive or negative depending on whether realized volatility
is greater or less than implied volatility. The local volatility P&L is positive or
negative depending on whether realized volatility is greater or less than shortterm local volatility.
Which of these hedged P&Ls is closer to zero on average, i.e. which one produces a better hedging strategy?
Combining Equation 9.6 and Equation 9.7, we see that in a local volatility
model,
1
2
2
2
δΠ BS = δΠ loc – ε × δ S =  Γ loc S [ σ R – σ ( S, δ t ) ]δ t – ε × δ S
2 Eq.9.8 This is our formula for the P&L of a BlackScholes hedging strategy in a world
where the skew is negative.
Let’s analyze this as Crepey does. There are two sources of BlackScholes contributions to P&L in Equation 9.8, one from the gamma term which is quadratic and nondirectional, and depends on the volatility mismatch, and one 4/18/08 smilelecture9.fm E4718 Spring 2008: Derman: Lecture 9:Patterns of Volatility Change Page 16 of 16 Copyright Emanuel Derman 2008 from the hedging mismatch which is directional and depends on the sign of
δS . 4/18/08 Crepey discusses four different market regimes, grouped along two axes, volatility and direction, as displayed in Table 5. Indexes can move up or down, with
high or low realized volatility compared to instantaneous local volatility.
For volatile down markets (a fast sellof, as Crepey calls it), both terms in
Equation 9.8 are positive, and the errors to δΠ BS are additive. The BlackScholes P&L differs from zero (the perfect hedge value) due to two additive
contributions.
For nonvolatile up markets (slow rise), both terms are negative and the same
is true.
For slow selloffs or fast rises, the two terms in Equation 9.8 have opposite
signs, and the hedging errors tend to cancel.
TABLE 5. Types of Markets: Equity index markets have the
characteristics of the yellow cells. Volatility
volatile nonvolatile up σ R > σ ( S, δ t ) , δ S > 0 σ R < σ ( S, δ t ) , δ S > 0 down σ R > σ ( S, δ t ) , δ S < 0 σ R < σ ( S, δ t ) , δ S < 0 Direction Therefore, the BlackScholes hedging strategy will perform worst in fast selloffs or slow rises. The localvolatility hedging strategy performs worst in slow
sellofs and fast rises.
Negatively skewed equity index markets are precisely characterized by slow
rises and fast selloffs. Therefore the BlackScholes hedging strategy is worse
than the local volatility strategy in these characteristic regimes. Crepey has
also backtested the hedging of actual options to show that the P&L of a hedged
portfolio has less variance under the local volatility hedging strategy.
Therefore, Crepey argues, for negatively skewed index markets, local volatility
models have hedge ratios that, on average, tend to work better than pure BlackScholes hedge ratios. smilelecture9.fm ...
View
Full
Document
This note was uploaded on 01/31/2011 for the course PSYCH 121 taught by Professor John during the Summer '10 term at UC Davis.
 Summer '10
 john

Click to edit the document details