GS-Implied Trinomial Trees
GS-Implied Trinomial Trees
GS-Implied Trinomial Trees
February 1996
Implied Trinomial Trees of
the Volatility Smile
Emanuel Derman
Iraj Kani
Neil Chriss
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
This material is for your private information, and we are not soliciting any action based upon it. This report is not to
be construed as an offer to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer
or solicitation would be illegal. Certain transactions, including those involving futures, options and high yield
securities, give rise to substantial risk and are not suitable for all investors. Opinions expressed are our present
opinions only. The material is based upon information that we consider reliable, but we do not represent that it is
accurate or complete, and it should not be relied upon as such. We, our affiliates, or persons involved in the
preparation or issuance of this material, may from time to time have long or short positions and buy or sell securities,
futures or options identical with or related to those mentioned herein.
This material has been issued by Goldman, Sachs & Co. and/or one of its affiliates and has been approved by
Goldman Sachs International, regulated by The Securities and Futures Authority, in connection with its distribution
in the United Kingdom and by Goldman Sachs Canada in connection with its distribution in Canada. This material is
distributed in Hong Kong by Goldman Sachs (Asia) L.L.C., and in Japan by Goldman Sachs (Japan) Ltd. This
material is not for distribution to private customers, as defined by the rules of The Securities and Futures Authority
in the United Kingdom, and any investments including any convertible bonds or derivatives mentioned in this
material will not be made available by us to any such private customer. Neither Goldman, Sachs & Co. nor its
representative in Seoul, Korea is licensed to engage in securities business in the Republic of Korea. Goldman Sachs
International or its affiliates may have acted upon or used this research prior to or immediately following its
publication. Foreign currency denominated securities are subject to fluctuations in exchange rates that could have an
adverse effect on the value or price of or income derived from the investment. Further information on any of the
securities mentioned in this material may be obtained upon request and for this purpose persons in Italy should
contact Goldman Sachs S.I.M. S.p.A. in Milan, or at its London branch office at 133 Fleet Street, and persons in Hong
Kong should contact Goldman Sachs Asia L.L.C. at 3 Garden Road. Unless governing law permits otherwise, you
must contact a Goldman Sachs entity in your home jurisdiction if you want to use our services in effecting a
transaction in the securities mentioned in this material.
Note: Options are not suitable for all investors. Please ensure that you have read and understood the
current options disclosure document before entering into any options transactions.
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
SUMMARY
In options markets where there is a significant or persis-
tent volatility smile, implied tree models can ensure the
consistency of exotic options prices with the market
prices of liquid standard options.
Table of Contents
INTRODUCTION .........................................................................................1
IMPLIED THEORIES ...................................................................................2
Implied Binomial Trees.................................................................. 3
TRINOMIAL TREES ................................................................................... 5
Implied Trinomial Trees .................................................................6
CONSTRUCTING THE IMPLIED TRINOMIAL TREE ......................................7
A DETAILED EXAMPLE.............................................................................9
What Can Go Wrong? .................................................................. 11
Constructing the State Space for the Implied Trinomial Tree ..... 15
APPENDIX A: Constructing Constant-Volatility Trinomial Trees ..... 19
APPENDIX B: Constructing Skewed Trinomial State Spaces .............21
0
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
INTRODUCTION Binomial trees are perhaps the most commonly used machinery for
options pricing. A standard Cox-Ross-Rubinstein (CRR) binomial tree
[1979] consists of a set of nodes, representing possible future stock
prices, with a constant logarithmic spacing between these nodes.
This spacing is a measure of the future stock price volatility, itself
assumed to be constant in the CRR framework. In the continuous
limit, a CRR tree with an “infinite” number of time steps to expira-
tion represents a continuous risk-neutral evolution of the stock price
with constant volatility. Option prices computed using the CRR tree
will converge to the Black-Scholes continuous-time results [1973] in
this limit.
1
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
The use of trinomial trees for building implied models has been sug-
gested by Dupire [1994]. The extra parameters in trinomial trees
give us the freedom to choose the price (that is, the location in “price
space”) at each node in the tree. This freedom to pre-specify the “state
space” can be quite advantageous if used judiciously. After an appro-
priate choice of the state space has been made, the transition proba-
bilities can be iteratively calculated to ensure that all European
standard options (with strikes and maturities coinciding with the
tree nodes) will have theoretical values which match their market
prices. We will show the equations to perform the iteration are very
similar to those derived for implied binomial trees.
IMPLIED THEORIES Implied theories assume that the stock (or index) price follows a pro-
cess whose instantaneous (local) volatility σ ( S, t ) varies only with
spot price and time4. Under this assumption, since all uncertainty in
the local volatility is derived from uncertainty in the stock price, we
can hedge options using the stock, and so, as in the traditional Black-
3. To be honest, we point out that no one really knows all the market options prices
needed to find the price and transition probabilities at every tree node. In practice,
there are market quotes for a discrete set of commonly traded strikes and expira-
tions, and market participants interpolate or extrapolate implied volatilities to other
points.
4. This process is an extension of constant volatility lognormal process, and is
described by the stochastic differential equation:
dS
------ = r ( t )dt + σ ( S, t )dZ
S
where dZ is the standard Wiener process, r ( t ) is the risk-free rate of interest at
time t , and σ ( S, t ) is the local volatility assumed to depend only on the future time
t and future spot price S .
2
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Implied Binomial Trees An implied binomial tree is a discrete version of a continuous evolu-
tion process that fits current options prices, in much the same way as
the standard CRR binomial tree is a discrete version of the Black-
Scholes constant volatility process5. Figure 1 shows schematic repre-
sentation of an implied binomial tree compared to a standard CRR
tree. The node spacing is constant throughout a standard CRR tree,
whereas in an implied binomial tree it varies with market level and
time, as specified by the local volatility function σ ( S, t ) .
Derman and Kani [1994] show how to construct the implied binomial
tree inductively. Figure 2 depicts the parameters to be determined in
moving from level n to level n+1 starting from an already known
5. For a review of implied binomial tree models, also see Chriss [1996a].
3
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
pi o Si+1
(n,i) si o Fi , Ci
o Si
level n n+1
time tn tn+1
∆t
representative stock price si at time tn. All node prices and transition
probabilities up to the nth level of the tree at time tn are assumed to
be known at this stage. We want to determine the node prices at the
(n+1)th level of the tree at time tn+1 along with transition probabili-
ties for moving from level n to level n+1 of the tree. There are 2n+1
unknown parameters; n+1 node values Si and n transition probabili-
ties p, to be determined from the known values of n forward contracts
Fi whose delivery date is tn+1 with delivery price Si , and n options Ci
expiring at tn+1 with strike Si. Derman and Kani [1994] provide the
detailed algorithm that fixes the unknown parameters from the
known prices6.
Aside from the choice of the central trunk, the implied binomial tree
is uniquely determined from forward and option prices. As mentioned
4
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
TRINOMIAL TREES Trinomial trees provide another discrete representation of stock price
movement, analogous to binomial trees8. Figure 3 illustrates a single
time step in a trinomial tree. The stock price at the beginning of the
time step is S0. During this time step the stock price can move to one
of three nodes: with probability p to the up node, value Su; with prob-
ability q to the down node, value Sd; and with probability 1 – p – q to
the middle node, value Sm. At the end of the time step, there are five
unknown parameters: the two probabilities p and q, and the three
node prices Su, Sm and Sd.
o Su
p
1-p-q
S0 o o Sm
q
o Sd
8. We remind the reader that both trinomial and binomial trees approach the same
continuous time theory as the number of periods in each is allowed to grow without
limit. Despite their limiting similarity, one kind of tree may sometimes be more con-
venient than another.
5
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
pS u + qS d + ( 1 – p – q )S m = F 0 (EQ 1)
If the stock price volatility during this time period is σ , then the node
prices and transition probabilities satisfy:
p ( S u – F 0 ) 2 + q ( S d – F 0 ) 2 + ( 1 – p – q ) ( S m – F 0 ) 2 = F 02 σ 2 ∆t + O ( ∆t ) (EQ 2)
where Ο(∆t) denotes terms of higher order than ∆t. Different discreti-
zations of risk-neutral trinomial trees have different higher order
terms in Equation 2.
Of the five parameters needed to fix the whole tree, Equations 1 and
2 provide only two constraints, and so we have three more parame-
ters than are necessary to satisfy them. By contrast, for implied bino-
mial trees, all unknown parameters were determined by the
constraints. As a result, we can construct many “economically equiv-
alent” trinomial trees which, in the limit as the time spacing ∆t
becomes very small, represent the same continuous theory. Appendix
A discusses a few different ways for building constant volatility trino-
mial trees. When volatilities are not constant, a common method is to
choose the stock prices at every node and attempt to satisfy the two
constraints through the choice of the transition probabilities. This
method of initially choosing the state space of prices for the trinomial
tree, and then solving for the transition probabilities, is familiar in
most applications of the finite-difference method. We must make a
judicious choice of the state space in order to insure that the transi-
tion probabilities remain between 0 and 1, a necessary condition for
the discrete world represented by the tree to preclude arbitrage.
Implied Trinomial Trees Figure 4 gives schematic representations for both standard and
implied trinomial trees.
6
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
CONSTRUCTING THE Suppose that we have already fixed the state space of the implied tri-
IMPLIED TRINOMIAL TREE nomial tree. Figure 5 shows the nth and (n+1)th levels of the tree. We
will use induction to infer the transition probabilities pi and qi for all
tree nodes (n,i) at each tree level n. Our notation and treatment fol-
lows the Derman and Kani [1994] binomial tree construction.
p i S i + 2 + ( 1 – p i – q i )S i + 1 + q i S i = F i (EQ 3)
7
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
If we set the strike K to the value Si+1, the stock price at the node
(n+1,i+1), then we can rearrange the terms and use Equation 3 to
write the call price in terms of known Arrow-Debreu prices, known
stock prices, known forward prices, and a contribution from up-tran-
sition probability pi to the first in-the-money node:
2n
e r∆t C ( S i + 1, t n + 1 ) = λi pi ( Si + 2 – Si + 1 ) + ∑ λ j ( F j – Si + 1 ) (EQ 5)
j = i+1
8
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
2n
e r∆t C ( S i + 1, t n + 1 ) – ∑ λ j ( F j – Si + 1 )
j = i+1
p i = ------------------------------------------------------------------------------------------------- (EQ 6)
λi ( Si + 2 – Si + 1 )
Using Equation 3 we can solve for the down transition probability qi:
Fi – pi ( Si + 2 – Si + 1 ) – Si + 1
q i = -----------------------------------------------------------------
- (EQ 7)
Si – Si + 1
i–1
Fi + qi ( Si + 1 – Si ) – Si + 1
p i = ----------------------------------------------------------- (EQ 9)
Si + 2 – Si + 1
We can now use Equation 2 to find the local volatility σ at each node.
9
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
152.85 152.85
B
123.63 123.63 123.63
A
100.00 100.00 100.00 100.00
65.43 65.43
52.92
10
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
= 14.6 %
The difference between the 14.6% implied local volatility and the
15% implied volatility assumed for this option arises from the higher
order terms in Equation 2, and will vanish as the time spacing
approaches zero.
What Can Go Wrong? The transition probabilities in Equations 6-9 for any node must lie
between 0 and 1, otherwise the implied tree allows riskless arbi-
trages which are inconsistent with rational options prices. For
implied trinomial trees, there are two possible causes for negative
probability at a node. First, the forward price Fi of a node (n,i) may
fall outside the range Si to Si+2 as illustrated in Figure 8. In that case
the forward condition of Equation 3 cannot be satisfied with all tran-
sition probabilities lying between 0 and 1.
11
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
0 1 2 3
time
(years)
0.421 0.416
0.544
0.134 0.278
0.437
0.017
Arrow-Debreu price tree:
0.083 0.133
nodes show λi
0.304 0.292 0.246
1.000 0.443 0.295 0.212
0.042 0.030
0.017
0.171 0.169
0.201
12
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
13
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
FIGURE 9.For the skew of 1 percentage point for every 10 strike points,
the 15% constant-volatility trinomial tree has negative probabilities at
nodes A, B and C. These probabilities, shown in larger type, have
been overwritten while maintaining the forward price condition.
0 1 2 3
time
(years)
A
up-transition probability tree: 0.371
nodes show pi 0.260 0.248
0.334 0.334 0.317
B
0.512 0.371
C
0.371
14
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
1 Fi – Si + 1 Fi – Si 1 Si + 2 – Fi
p i = --- ---------------------------
- + ---------------------- and q i = --- ----------------------
2 Si + 2 – Si + 1 Si + 2 – Si 2 Si + 2 – Si
1 Fi – Si 1 Si + 2 – Fi Si + 1 – Fi
p i = --- ---------------------- and q i = --- ---------------------
- + ----------------------
2 Si + 2 – Si 2 Si + 2 – Si Si + 1 – Si
Constructing the State Regular state spaces with uniform mesh sizes are usually adequate
Space for the Implied for constructing implied trinomial tree models when implied volatil-
Trinomial Tree ity varies slowly with strike and expiration. But if volatility varies
significantly with strike or time to expiration, it may be necessary to
choose a state space whose mesh size (or node spacing) changes sig-
nificantly with time and stock level.
The implied trinomial tree model constructed using this skewed state
space has no negative probabilities, fits the option market prices
accurately, and generates reasonably smooth values for local volatil-
ity at different stock and time points.
One way to construct trinomial state spaces with proper skew and
term structure is to build it in the following two stages:
• First, assume all interest rates (and dividends) are zero and build
a regular trinomial lattice with constant time spacing ∆t and loga-
rithmic level spacing ∆S. Any constant volatility trinomial tree
corresponding to a typical market implied volatility (see Appendix
A) is an example of this type of lattice. Then modify ∆t at different
times and, subsequently, ∆S at different stock levels until the lat-
tice captures the basic term-and skew- structures of local volatility
in the market.
15
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
FIGURE 10.A skewed choice for the state space of the implied
trinomial tree model for the second example.
0 1 2 3
time
(years) 166.05
143.26 143.26
58.56 58.56
39.50
FIGURE 11.For the skew of 1 percentage point for every 10 strike points,
the skewed trinomial tree has no negative probabilities. The resulting
local volatilities at different nodes are reasonably smooth.
0 1 2 3
time
(years)
up-transition probability tree: 0.285
nodes show pi 0.357 0.345
0.372 0.372 0.371
0.370 0.369
0.338
16
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
We must point out that, for a fixed number of time and stock price
levels, it may be impossible to avoid all negative probabilities, no
matter what choice we make for the state space. As long as our choice
does not violate the forward price condition at any node, we can over-
write the option prices which produce negative probabilities. In this
way, even though we give up fitting the option price at some of the
implied tree nodes, we fit the forward prices with transition probabil-
ities which lie between 0 and 1 at every node. Generally, the less
overwriting we have to do in our implied tree, the better it will fit the
smile.
FIGURE 12. A schematic construction for the state space of the implied
trinomial tree model: (a) build a regular trinomial lattice with equal
time and price steps; (b) modify different time and then price steps in
the lattice; (c) grow the lattice along the forward interest rate curve.
9. It may not be necessary to grow the nodes precisely along the forward rate curve.
Any sufficiently large growth factor which removes forward price violations of the
types described in Figure 8 will be sufficient.
17
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
18
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
APPENDIX A: Constructing This appendix provides several methods for constructing constant
Constant-Volatility volatility trinomial trees that can serve as initial state spaces for
Trinomial Trees implied trees. The different methods described here will all converge
to the same theory, i.e the constant-volatility Black-Scholes theory, in
the continuous limit. In this sense, we can view them as equivalent
discretizations of the constant volatility diffusion process. Figure 13
shows two common methods for building binomial trees. There are in
general an infinite number of (equivalent) binomial trees, all repre-
senting the same discrete constant volatility world. This is due to a
freedom in the choice of overall growth of the price at tree nodes (not
to be confused with the stock’s risk-neutral growth rate). If we multi-
ply all the node prices of a binomial tree by some constant (and rea-
sonably small) growth factor, we will end up with another binomial
tree which has different (positive) probabilities but represents the
same continuous theory. The familiar Cox-Ross-Rubinstein (CRR)
binomial tree has the property that all nodes with same spatial index
have the same price. This makes CRR tree look regular in both spa-
tial and temporal directions. The Jarrow-Rudd (JR) binomial tree10
has the property that all probabilties are equal to 1/2. This property
makes the JR tree a natural discretization for the Brownian motion.
The JR tree does not grow precisely along the forward risk-free inter-
est rate curve, but we can just as easily construct binomial trees
which have this property11.
2
Su = Se σ ∆t Su = Se ( r – σ ⁄ 2 )∆t + σ ∆t
Sd = Se – σ ∆t 2
Sd = Se ( r – σ ⁄ 2 )∆t – σ ∆t
p = (SF - Sd) /(Su - Sd) != 1/2 p = (SF - Sd) /(Su - Sd) = 1/2
Su Su
S
Sd S Sd
19
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
(a) Combining two steps of (b) Combining two steps of (c) Equal-probability tree
a CRR binomial tree a JR binomial tree
Se σ 2 ⁄ 2 )∆t + σ 2 ∆t S = ( r – σ 2 ⁄ 2 )∆t + σ 3 ∆t ⁄ 2
Su = 2∆t
Su = Se ( r – σ u Se
Sm = S 2 ⁄ 2 )∆t
Sm= Se ( r – σ
2 ⁄ 2 )∆t
Sm = Se ( r – σ
Sd = Se –σ 2∆t
2 ⁄ 2 )∆t – σ 3 ∆t ⁄ 2
Sd = Se ( r – σ
2 ⁄ 2 )∆t – σ 2 ∆t
e r∆t ⁄ 2 – e –σ ∆t ⁄ 2
2 Sd = Se ( r – σ
p = ----------------------------------------------
-
e σ ∆t ⁄ 2 – e – σ ∆t ⁄ 2
p = 1/4 p = 1/3
2
e σ ∆t ⁄ 2 – e r∆t ⁄ 2
q = ----------------------------------------------
- q = 1/4 q = 1/3
e σ ∆t ⁄ 2 –e – σ ∆t ⁄ 2
Su Su
Su
S p p p
Sm Sm S Sm
S q
q q
Sd Sd Sd
12. In a recombining constant volatility trinomial tree Su, Sm, and Sd have the gen-
eral form S u = Se π∆t + φσ ∆t, S m = Se π∆t and S d = Se π∆t – φσ ∆t for φ > 1 and any rea-
sonable value of π.
20
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Assume that interest and dividend rates are zero for now. Consider
the term-structure case first. Here local volatility is some function of
time σ ( t ) . We can introduce the notion of scaled time t̃ as
t̃
∫
t = c σ 2 ( u ) du
0
for some constant c. Differentiating both sides of this equation we can
write an equivalent non-linear equation describing t̃ in terms of t:
t
∫
1 1
t̃ = --- --------------------- du
c σ 2 ( t̃ ( u ) )
0
Using the scaled time in place of standard time transforms the stock
evolution process to a constant volatility (Black-Scholes) process15.
We can choose the constant c so that the rescaled and physical times
coincide at some fixed future time T, e.g. t̃ ( T ) = T . In this case
T T
∫ ∫
1 1
c = T ⁄ σ 2 ( u ) du = --- --------------------- du
T σ ( t̃ ( u ) )
2
0 0
13. If Σ = Σ(T) is the implied volatility for expiration T, then local volatility at time
t=T is given by the relation σ2(T) = d[TΣ2(T)]/dT.
14. If Σ = Σ0 + b(K-K0) is implied volatility for strike price K, then the local volatility
at level K in the vicinity of K0 is roughly given by the relation σ = Σ0 + 2b(K-K0)
15. Define a new stock price variable S̃ by the relation S̃ ( t ) = S ( t̃ ) and also a new
Brownian motion Z̃ by t t
1
------Z̃ c ∫ σ ( u ) du = ∫ σ ( u ) dZ ( u ) .
2
c
0 0
21
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
∑ -------------
1
T -
i=1
σ ( t˜ )
2
i
- ; k = 1, ..n
t̃ k = ---------------------------
n
∑
1
--------------
i=1
σ 2 ( t˜ )
i
∫
1
S̃ = S 0 exp c --------------- dx
xσ ( x )
S0
for some constant c. The scaled stock price has a constant volatility
equal to c17. A reasonable discrete representation of scaled stock
price movements can be given by a constant volatility tree. Inverting
this equation, we can convert the discrete nodal values of S̃ to dis-
crete values of S . In the resulting S -tree the spacing between nodes
varies with the level corresponding to the similar variation in local
volatility. We can choose the constant c to represent the at-the-money
or some other typical value of local volatility. For any fixed time
period, if S̃ i denote the nodal values of scaled stock price, the corre-
sponding stock price values can be found by solving the discrete ver-
sion of the above equation. For nodes which lie above the central node
this gives:
1 S̃ k
S k = S k – 1 + --- σ ( S k – 1 )S k – 1 log -----------
c S̃ k – 1
22
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
Again we can set the constant c to the at-the-money or any other rea-
sonable value of local volatility19.
S γ
σ ( S, t ) = σ 0 ( t ) -----
S0
23
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
REFERENCES
Cox, J.C., S.A. Ross and M. Rubinstein (1979). Option Pricing: A Sim-
plified Approach, Journal of Financial Economics 7, 229-263.
24
Goldman
Sachs
QUANTITATIVE STRATEGIES RESEARCH NOTES
25