# The BlackScholesMertonformulaSome Consequences

Shared by:
Categories
-
Stats
views:
3
posted:
7/9/2010
language:
English
pages:
2
Document Sample

```							                                                                            Volatility
where P (t, T ) = exp − tT r(u)du     is the zero coupon bond price and     What matters is the average (or integrated) (squared) volatility during
the life of the option. “Future volatility, not past volatility”.
1    T
2
σAV =             σ 2 (u)du.
T −t t                                      If volatility is stochastic, σ(t, ω), then the random variable tT σ 2(u, ω)du
still plays an important role for option pricing, although things are
This is (sometimes) called the Black/Scholes/Merton formula.                quite as nice as in the r-case.

Not too hard to prove with martingale methods; hard to guess with           If σ is exogenous to the stock price (driven by another BM) then by
PDEs.                                                                       conditioning:
2
call(t) =   callBSM (. . . , σAV = x, . . .)φσ 2 (x)dx,
Deterministic changes may not be what we ultimately want, but it’s                                                                  AV

a ﬁrst step. “Shades of the real world” & gives useful information.         where φσ 2 denotes the density of T 1 tT σ 2(u, ω)du. Expansions lead
−t
AV
to (approximate) formulas. (Hull & White 1987.)
2                                                                         4

The Black/Scholes/Merton formula & Some Consequences                       When r and σ are constant        original B/S formula. Time-varying
dividend yields: DIY (relevant in currency models).
Exercise 4 on WN 9 extends base-case B/S to time-varying (but
deterministic) interest rate r(t) and volatility σ(t).                      Interest rates
If there is a term structure of interest rates, then in place of r in
The arbitrage-free call-option price is                                     original B/S formula you should plug in the zero coupon rate with
                                                maturity equal to the time to expiry of the option:
S(t)

ln( P (t,T )K ) + 1 σAV (T − t) 
2
2
call(t) = S(t)Φ                  √                                     r    y(t, T ) (since by deﬁnitionP (t, T ) = exp(−y(t, T )(T − t)))

σAV T − t


                                         With stochastic interest rates, the ZC yield y(t, T ) can also be plugged
S(t)

ln( P (t,T )K ) − 1 σAV (T − t) 
2
2
in. A good approximation for stocks. Can be theoretically supported
−KP (t, T )Φ                  √               ,

σAV T − t                     when interest rates are Gaussian & independent of the stock.

1                                                                         3
The LHS can be estimated from data by summing squared log-returns
over the observation period [Ti; Ti]. This is called realized volatility

If the volatility path is known to market participants, then the BSM-
analysis tells us that the (Black/Scholes) implied volatility of an op-
T
tion (after scaling) is T i+1 σ 2 (u, ω)du. This can be recorded/observed
i

This leads to the empirical/econometric question: Is implied volatil-
ity a good forecast for realized volatility? In particular, does implied
volatility contain information that can’t be extracted from past/historical
data?

6

Other Spin-Oﬀ: Implied vs. Realized Volatility
Empirical analysis: Record data     time series of realized volatilities,
If σ is stochastic and
σAV (tk ) and corresponding implied volatilities σIM P (tk ).
dS(t) = S(t)µ(t)dt + σ(t, ω)S(t)dW P ,
then with Y = ln S we have                                                    Run regression (possibly in logs)
dY = (µ − σ 2/2)dt + σ(t, ω)dW P                                         σAV (tk ) = α0 + α1σIM P (tk ) + (tk+1)

Now look at some [Ti ; Ti+1] and suppose it has been split into n pieces      If α1 = 0, then implied volatility contains some information.
(intermediate points tj ).                                                    If (α0, α1) = (0, 1) then implied volatility is an unbiased forecast.
Include eg. α2σAV (tk−1 ); test if α2 = 0.
Then from our analysis of quadratic variation we know that
n→∞      Ti+1                           Results: There is some information content in implied volatility that
(Y (tj+1 ) − Y (tj ))2 −→           σ 2 (u, ω)du
j                               Ti                              historical volatility does not capture. Beware of overlapping data!
Reference: Christensen & Prabala (JFE 1998).
(All objects are random variables; convergence is in L2-sense.)
5                                                                           7

```