Академический Документы
Профессиональный Документы
Культура Документы
)
September October Nov
2 Short-squeeze in Volkswagen: October 2008
avellaneda.indd 93 27/5/09 11:35:00
94 Risk June 2009
CUTTING EDGE. MARKET RISK
borrowing can be measured using leveraged exchange-traded
funds (ETFs) tracking nancial stocks (which were extremely
hard to borrow in autumn 2008).
The model
We assume that under the physical measure the hard-to-borrow
stock S
t
and buy-in rate i
t
satisfy the system of coupled equations:
dS
t
S
t
odW
t
+ yi
t
dt ydN
i
t
t ( ) (1)
dX
t
xdZ
t
+ o X X
t
( )dt +
dS
t
S
t
, X
t
ln i
t
/ i
0
( ) (2)
where dN
i
(t) denotes the increment of a standard Poisson process
with intensity i
t
over the interval (t, t + dt); the parameters o and
y are respectively the volatility of the continuous part and the
price elasticity of demand due to buy-ins; and W
t
is a standard
Brownian motion. Equation (2) describes the evolution of the
logarithm of the buy-in rate; x is the volatility of the rate, Z
t
is a
Brownian motion independent of W
t
, X
_
is a long-term equilib-
rium value for X
t
, o is the speed of mean-reversion and couples
the change in price with the buy-in rate.
We assume that > 0; in particular, X
t
= ln(i
t
) is positively cor-
related with price changes, introducing a positive feedback
between increases in buy-ins (hence in short interest in the stock)
and price.
Equations (1) and (2) describe the evolution of the stock price
across an extended period of time. One can think of a diusion
process for the stock price, which is punctuated by jumps occur-
ring at the end of the trading day, the magnitude and frequency
of the latter being determined by i
t
. Fluctuations in i
t
represent
the fact that a stock may be dicult to borrow one day and easier
another. In this way, the model describes the dynamics of the
stock price as costs for stock-loan vary. Short squeezes can be seen
as events associated with large values of i
t
, which will invariably
exhibit price spikes (rallies followed by a steep drop).
The cost of shorting: buy-ins and efective dividend yield
Option market-makers need to hedge by trading the underlying
stock, both on the long and short side, with frequent adjustments.
However, securities that become hard to borrow are subject to
buy-ins as the rm needs to deliver shares according to the pres-
ently existing settlement rules. From a market- makers viewpoint,
a hard-to-borrow stock is essentially a security that presents an
increased likelihood of buy-ins.
Te prot or loss for a market-maker is aected by whether and
when their short stock is bought in and at what price. Generally,
this information is not known until the end of the trading day. To
model the economic eect of buy-ins, we assume that the traders
prot and loss from a short position of one share over a period (t,
t + dt) is:
P & L dS
t
yS
t
S
t
odW
t
+ i
t
ydt ( )
where Prob{ = 0} = 1 i
t
dt + o(dt) and Prob{ = 1} = i
t
dt +
o(dt). Tus, we assume that the trader who is short the stock does
not benet from the downward jump in equation (1) because they
are no longer short by the time the buy-in is completed. Te idea
is that the short trader takes an economic loss post-jump due to
the fact that their position was closed at the buy-in price.
Suppose then, hypothetically, that the trader was presented
with the possibility of renting the stock for the period (t, t + dt)
so that they can remain short and be guaranteed not to be bought
in. Te corresponding prot and loss would now include the neg-
ative of the downward jump, that is, yS
t
if the jump happened
right after time t. Since jumps and buy-ins occur with frequency
i
t
, the expected economic gain is i
t
yS
t
. It follows that the fair
value of the proposed rent is i
t
y per dollar of equity shorted. In
other words, i
t
y can be viewed as the cost-of-carry for borrowing
the stock.
Hence, we can interpret i
t
y as a convenience yield associated
with owning the stock when the buy-in rate is i
t
. Tis conven-
ience yield is monetised by holders of long positions lending
their stock out for one day at a time and charging the fee associ-
ated with the observed buy-in rate. Te convenience yield or
rent is mathematically equivalent to a stochastic dividend yield
that is credited to long positions and debited from holders of
short positions who enter into lending agreements. For traders
who are short but do not enter into such agreements, it is
assumed that stochastic buy-ins prevent them from gaining
from downward jumps.
We can therefore introduce an arbitrage-free pricing measure
associated with the physical process (1)(2), in which the rent, or
stock-nancing, i
t
y, cancels the drift component of the model
and the expected return is equal to the cost of carry. Under this
measure, our model takes the form:
dS
t
S
t
odW
t
+ rdt ydN
i
t
t ( ) (3)
where r is the instantaneous interest rate. Te absence of the
drift term i
t
y in this last equation is due to the fact that, under
an arbitrage-free pricing measure, the discounted price process
is a martingale.
It follows from equation (3) that the stock price in the risk-neu-
tral world can be written as:
S
t
S
0
e
rt
M
t
1 ( )
dN
t
t ( )
0
t
(4)
where the third factor represents the eects of buy-ins, and:
M
t
: exp oW
t
o
2
t
2
S
0
e
rT
E e
i
t
dt
0
T
[
i
t
dt
0
T
[
_
,
k
k!
1 y ( )
k
k
_
S
0
e
rT
E e
y i
t
dt
0
T
[
(5)
Tis gives a mathematical formula for the forward price in
terms of the buy-in rate and the constant y. Clearly, if there are no
jumps, the formula becomes classical. Otherwise, notice that the
avellaneda.indd 94 27/5/09 11:35:01
risk.net 95
dividend is positive and delivering stock into a forward contract
requires hedging with less than one unit of stock, renting it along
the way to arrive at one share at delivery. From equation (5), the
term structure of forward dividend yields (d
t
) associated with the
model is given by:
e
d
t
dt
0
T
E e
y i
t
dt
0
T
[
(6)
Option pricing for hard-to-borrow stocks
Put-call parity for European-style options states that:
C K,T ( ) P K,T ( ) = S 1 DT ( ) K 1 RT ( )
where P(K, T), C(K, T) represent respectively the fair values of a
put and a call with strike K and maturity T, S is the spot price and
R, D are respectively the simply discounted interest rate and divi-
dend rate. It is equivalent to:
C
pop
K,T ( ) P
pop
K,T ( ) = KRT DST
(7)
where P
pop
(K, T) = P(K, T) max(K S, 0) represents the pre-
mium-over-parity for the put, a similar notation applying to calls.
It is well known that put-call parity does not hold for hard-to-
borrow stocks if we enter the nominal rates and dividend rates in
equation (7). Te price of conversions in actual markets should
therefore reect this. A long put position is mathematically equiv-
alent to being long a call and short 100 shares of common stock,
but this will not hold if the stock is a hard-to-borrow stock. Te
reason is that shorting costs money and the arbitrage between
puts and calls on the same line, known as a conversion, cannot be
made unless there is stock available to short. Conversions that
look attractive, in the sense that:
C
pop
K,T ( ) P
pop
K,T ( ) < KRT DST (8)
may not result in a risk-free prot due to the fact that the crucial
stock hedge (short 100 shares) may be impossible to establish.
We quantify deviations from put-call parity by considering the
function:
d
imp
K,T ( )
C
pop
K,T ( ) P
pop
K,T ( ) KRT
ST
, 0 K
(9)
As a function of K, d
imp
(K, T) will be approximately at for low
strikes and will rise slightly for large values of K because puts
become more likely to be exercised. Te dividend yield for the
stock should correspond roughly to the level of d
imp
(K, T) for at-
the-money strikes. If we consider American-style options on divi-
dend-paying stocks or exchange-traded funds (for example, the
S&P 500), then the implied dividend curve will, in addition, be
lower for low strikes, reecting the fact that calls have an early-
exercise premium.
Te situation is quite dierent for hard-to-borrow stocks, as we
can see from gures 4 and 5. Two distinctions are important: (i)
the implied dividend curve d
imp
(K, T) for K - S is not equal to the
nominal dividend yield (which is zero, in the case of the stocks
that are displayed in the gures) instead, it has a positive value;
and (ii) the implied dividend curve d
imp
(K, T) also bends for low
values of the strike, suggesting that calls with low strikes should
have an early exercise premium.
Te rst feature a change in level in the implied dividend
curve has to do with the extra premium for being long puts in a
world where shorting stock is dicult or expensive. Since syn-
thetic puts cannot be manufactured by owning calls and shorting
stock, the nominal put-call parity does not hold. Instead, it is
replaced by a functional put-call parity, which expresses the rela-
tive value of puts and calls via an eective dividend rate. Indeed,
if we dene the at-the-money implied dividend yield D
*
(T) =
d
imp
(S, T), we obtain the new parity relation:
C
pop
K,T ( ) P
pop
K,T ( ) = KRT D
*
T ( )ST
According to our model, we have, from equation (9):
D
*
T ( )
1 e
d
t
dt
0
T
[
T
1 E e
y i
t
dt
0
T
[
1
]
1
1
1
T
(10)
which connects the implied dividend rate obtained from the
options markets to the buy-in rate process.
0.50
0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0
I
m
p
l
i
e
d
d
i
v
i
d
e
n
d
(
%
)
Strike ($)
0 5 10 15 20 25 30 35 40 45
Note: the trade date is January 10, 2008 and the expiry is January 17,
2009. The stock price is $5.81. The best fit constant dividend rate is
approximately 15%. Dendreon does not pay dividends
4 Implied dividend rates as a function of strike price for
options on Dendreon
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0
I
m
p
l
i
e
d
d
i
v
i
d
e
n
d
(
%
)
Strike ($)
0 50 100 150 200 250 300
Note: the dates are as in figure 4 and the stock price is $80.30. The
best fit dividend rate (associated with at-the-money options) is 5.5%.
VMWare does not pay dividends
5 Implied dividend rates for VMWare
avellaneda.indd 95 27/5/09 11:35:02
96 Risk June 2009
CUTTING EDGE. MARKET RISK
Te option market predicts dierent borrowing rates over time
for any given stock, through variations in the interest rate quoted
by clearing rms, and by conversion-reversals quoted by option
market-makers. Te latter approach suggests dierent implied
dividends per option series, that is, it contains market expecta-
tions of the varying degree of diculty of borrowing a stock in
the future. We can use the model (1)(2) and equation (10) to
calculate a term structure of eective dividends (or, equivalently,
short rates) that could be calibrated to any given stock. To gener-
ate such a term structure, we simulate paths of i
t
, 0 < t < T
max
and
calculate the discount factors by Monte Carlo. Figure 6 shows a
declining term structure, which is typical of most stocks. Tis
decay represents the fact that stocks rarely remain hard to borrow
over extremely long time periods.
If we make the approximation that i
t
is independent of W
t
, in
the sense that:
dX
t
xdZ
t
+ o X X
t ( )
dt +
t
dt dN
t
t ( )
( )
the model becomes more tractable and we obtain semi-explicit
pricing formulas for European-style puts and calls as series expan-
sions by separation of variables. To see this, we dene the
weights:
H n,T ( ) Prob dN
i
t
t ( ) n
0
T
[
E e
i
t
dt
0
T
[
i
t
dt
0
T
[
_
,
n
n!
(11)
Denote by BSCall(s, t, k, r, d, o) the Black-Scholes value of a
call option for a stock with price s, time to maturity t, strike price
k, interest rate r, dividend yield d and volatility o. We then have:
C S, K,T ( ) H n,T ( ) BSCall S 1 y ( )
n
,T, K, r, 0,
( )
n0
(12)
with a similar formula holding for European-style puts.
Notice that equation (4) can be viewed as the risk-neutral
process for a stock that pays a discrete dividend yS
t
with fre-
quency i
t
. Terefore, calls will be exercisable if they are deep
enough in-the-money. A heuristic explanation is that a trader
long a call and short stock would suer repeated buy-ins costing
more than the synthetic put forfeited by exercising. Unfortu-
nately, pricing an American-style call using the full model (3)
entails a high-dimensional numerical calculation, because the
number of jumps until time t:
dN
t
t ( )
0
t
U
t
dS
t
S
t
r p
t
( )dt + rdt fdt, e 2, 2 { (13)
where r is the benchmark funding rate (Libor or Fed Funds), f is
8
9
10
11
12
13
14
15
16
T (years)
D
*
(
T
)
(
%
)
0 0.5 1.0 1.5 2.0
Note: D*(T) for
0
= 15, = 0.01, = 30, = 0.5
6 Term structure of effective dividend rates
0.18
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0
I
m
p
l
i
e
d
d
i
v
i
d
e
n
d
(
%
)
Strike
50 60 70 80 90 100 110 120 130 140 150
Note: we assume that the stock price is $100, = 0.50, = 1.00,
0
= 50, T = 0.5yrs, = 0.03, r = 10%. The effective dividend rate is
d
imp
(100, T) = 14%. For low strikes, the drop in value is related to
the early exercise of calls, a feature unique to hard-to-borrow
stocks. For high strikes, the broad increase corresponds to the
classical early exercise property of in-the-money puts
7 Theoretical implied dividend yield d
imp
(K, T) generated
by the model
avellaneda.indd 96 27/5/09 11:35:03
risk.net 97
the expense ratio or management fee and:
t
t
if 0
0 if 0
(14)
Tus p
t
is the instantaneous (annualised) rent that is associated
with shorting the underlying stock. We can view p
t
as a proxy
for yi
t
, the expected shortfall for a short-seller subject to buy-in
risk, or the fair reduced rate associated with shorting the
underlying asset.
We obtain:
dU
t
2 ( )
U
t
2 ( )
+
dU
t
2 ( )
U
t
2 ( )
= 2 r
t
( ) f ( )dt
which implies that:
t
dt
dU
t
2 ( )
U
t
2 ( )
dU
t
2 ( )
U
t
2 ( )
2 f 2r ( )dt
2
(15)
Tis suggests that we can use daily data on leveraged ETFs to
estimate the cost of borrowing the underlying stock.
For the empirical analysis, we used dividend-adjusted closing
prices from the PowerShares Ultrashort Financial ETF (SKF) and
the PowerShares Ultralong Financial ETF (UYG). Te underly-
ing ETF is the Barclays Dow Jones Financial Index ETF. Using
historical data, we calculated the right-hand side of equation (15),
which we interpret as corresponding to daily sampling, with dt =
1/252, r = 3-month Libor and f = 0.95%, the expense ratio of
SKF and UYG advertised by Powershares. Te results of the sim-
ulation are seen in gure 8.
We see that p
t
, the cost of borrowing, varies in time and can
change quite dramatically. In gure 8, we consider a 10-day mov-
ing average of p
t
to smooth out the eect of volatility and end-of-
day marks. Te data shows that increases in borrowing costs, as
implied from the leveraged ETFs, began in late summer 2008
and intensied in mid-September, when Lehman Brothers col-
lapsed and the US Securities and Exchange Commission (SEC)
ban on shorting 800 nancial stocks was implemented (the latter
occurred on September 19, 2008). Notice that the implied bor-
rowing costs for nancial stocks remain elevated subsequently,
despite the fact that the SEC ban on shorting was removed in
mid-October. Tis calculation may be interpreted as exhibiting
the variations of i
t
(or yi
t
) for a basket of nancial stocks. For
instance, if we assume that the elasticity remains constant (for
example, at 2%), the buy-in rate will range from a low number
(for example, i = 1, or one buy-in a year) to 50 or 80, correspond-
ing to several buy-ins a week.
Conclusion
In the past, attempts have been made to understand option pric-
ing for hard-to-borrow stocks using models that do not take
into account price dynamics. Tis approach leads to a view of
put-call parity that is at odds with the functional equilibrium
(steady state) evidenced in the options markets, in which put
and call prices are stable and yet naive put-call parity does not
hold. Te point of this article has been to show how dynamics
and pricing are intertwined. Te notion of eective dividend is
the principal consequence of our model, which also obtains a
term structure of dividend yields. Reasonable parametric choices
lead to a term structure that is concave down, a shape frequently
seen in real option markets. Te model also reproduces the
(American-style) early exercise features, including early exercise
of calls, which cannot happen for non-dividend-paying easy-to-
borrow stocks. N
Marco Avellaneda is professor of mathematics at New York University.
Mike Lipkin is adjunct professor of finance at Columbia University and
managing member of Katama Trading. They would like to thank the
referees for many helpful and insightful comments, Sacha Stanton of
Modus Incorporated for assistance with options data and Stanley Zhang
for exciting discussions on leveraged exchange-traded funds. Email:
marco.avellaneda@gmail.com, mike.katama@gmail.com
Amin K, 1993
Jump-difusion option valuation in
discrete time
Journal of Finance 48, pages
1,8331,863
Diamond D and R Verrecchia, 1987
Constraints on short-selling and
asset price adjustment to private
information
Journal of Financial Economics 18,
pages 277312
Dufe D, N Garleanu and
L Pedersen, 2002
Securities lending, shorting, and pricing
Journal of Financial Economics 66,
pages 307339
Evans R, C Geczy, D Musto and
A Reed, 2008
Failure is an option: impediments to
short selling and options prices
Review of Financial Studies, available
at http://rfs.oxfordjournals.org/cgi/
content/full/hhm083
Jones C and O Lamont, 2002
Short sale constraints and stock
returns
Journal of Financial Economics 66,
pages 207239
Natenberg S, 1998
Option volatility and strategies:
advanced trading techniques for
professionals
Probus, second edition
Nielsen L, 1989
Asset market equilibrium with short-
selling
Review of Economic Studies 56(3),
July, pages 467473
References
300
250
200
150
100
50
0
50
100
%
J
a
n
2
0
0
7
F
e
b
2
0
0
7
M
a
r
2
0
0
7
A
p
r
2
0
0
7
M
a
y
2
0
0
7
J
u
n
2
0
0
7
J
u
l
2
0
0
7
A
u
g
2
0
0
7
S
e
p
2
0
0
7
O
c
t
2
0
0
7
N
o
v
2
0
0
7
D
e
c
2
0
0
7
J
a
n
2
0
0
8
F
e
b
2
0
0
8
M
a
r
2
0
0
8
A
p
r
2
0
0
8
M
a
y
2
0
0
8
J
u
n
2
0
0
8
J
u
l
2
0
0
8
A
u
g
2
0
0
8
S
e
p
2
0
0
8
O
c
t
2
0
0
8
N
o
v
2
0
0
8
D
e
c
2
0
0
8
J
a
n
2
0
0
9
F
e
b
2
0
0
9
Note: the thin line corresponds to the daily values of the cost of
borrowing parameter
t
, in percentage points, estimated from
equation (16). The thick line is a 10-day moving average. Hard-to-
borrowness exceeds 100% in SeptemberOctober 2008 and remains
elevated until March 2009
8 The cost of borrowing the Barclays Dow Jones
Financial Index
avellaneda.indd 97 27/5/09 11:35:04