Brockhaus Oliver 03.12.2014

Introduction
Affine dividends
Homogenous dividend dynamics
Conclusion
Dividend Discount Approach to Equity Derivatives Modelling
Oliver Brockhaus
MathFinance AG
oliver.brockhaus@mathfinance.com
Cass Business School

London, 3 December 2014
oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 1 / 32

Introduction
Affine dividends
Conclusion
Agenda
1 Introduction
Research
Equity dynamics
Dividend discount models
Calibration
2 Affine dividends
Bos-Vandermark
Affine stock process
Affine dividend dynamics
3 Homogenous dividend dynamics
Indicator function
Korn-Rogers
Piecewise linear
Exponential
4 Conclusion
Summary
References

Introduction Research
Affine dividends Equity dynamics
Homogenous dividend dynamics Dividend discount models
Conclusion Calibration
Research
There are various directions of research on the topic of dividends:

Empirical properties, forecasting:
van Binsbergen et al. [vBBK10]
Dividends within equity dynamics:
Bos-Vandermark [BV02] (closed form formula with discrete dividends)
Overhaus et al. [OFK+ 02] and [OBB+ 07] (affine model)
Vellekoop, Nieuwenhuis [VN06] (model survey)
Dividend discount models:
Korn-Rogers [KR05] (dividend announcement)
Bernhart, Mai [BM12] (general framework)
Dividend derivatives modelling:
Tunaru [Tun14] (uncertain timing, cum dividend process)
Bühler et al. [BDS10] (discrete dividends with stochastic yield)

Equity dynamics
When modelling equity dynamics there are two popular ways of incorporating
dividends:
Continuous dividend payment stream, proportional to spot level S, namely
continuous dividend yield q:
dSt
= (rt − qt )dt + σt dWt
St
Discrete dividends with known future ex-dividend dates

0 < t1 < t2 < . . . , namely
Dtii = fi (Sti − )
with deterministic fi . In practice one often assumes

deterministic payments in the near future: fi = ci
proportional dividends for the distant future: fi (Sti − ) = ci Sti −
Positivity of S is preserved if fi is capped at Sti − or fi (x) ≤ x.

Dividend discount models
Dividend discount models assume that the stock price St can be viewed as the
present value of all future dividends, namely
X i
St = Dt (1)
ti >t
where Dti denotes the value discounted to t of the i-th dividend going ex at
time ti . Also, for any given stock model (St , t ≥ 0) with discrete dividends one
has
Sti − − Sti = Dtii (2)
In this presentation we address the following questions:

1 Given a stock model (St , t ≥ 0) with discrete dividends, are there
processes D i such that (1) holds?
2 Do these processes D i have ‘reasonable’ properties?
3 What is the dynamics of (St , t ≥ 0) if one starts with ‘reasonable’
processes D i ?

Calibration
Let Ft and Pt denote equity forward and discount factor, respectively. One has
the calibration condition
X i X i
Pt Ft = S0 − D0 = D0 (3)
ti ≤t ti >t
If a stock does not pay dividends then one may assume a single dividend
payment at infinity with value Dt = St at time t. Otherwise one can assume D0i
to be known for all i up to some index k.
With exponential growth thereafter, namely
D0i = D0k e −q∆(i−k) and ti = tk + ∆(i − k)
for all i ≥ k and some time increment ∆ one computes
S0 − i<k D0i
P
Pt Ft
q∆ = log P i
= log k−1 k−1
S0 − i≤k D0 Ptk Ftk
from (1) or (3) with t = 0. Growth parameter q can be interpreted as long

term dividend yield.

Introduction
Bos-Vandermark
Affine dividends
Conclusion
Bos-Vandermark
In a famous article Bos and Vandermark [BV02] present a variant of the Black
Scholes formula, namely

CBV (K , t) = Pt Black Pt−1 (S0 − Xtn ), K + Pt−1 Xtf , σ 2 t (4)
with deterministic functions

X (t − ti )+ i X ti ∧ t i
Xtn = D0 Xtf = D0
t t
ti ≤t ti ≤t
and volatility σ.
This formula is a good approximation for a European option price with

deterministic dividend drops and volatility σ in between. Additionally, the
formula is continuous both as valuation date moves across a dividend and
maturity moves across a dividend. The model as defined in their paper depends
on the maturity t of the European option.

Introduction
Bos-Vandermark
Affine dividends
Conclusion
Bos-Vandermark
We now fix t independently of the product, namely

X (T − ti )+ i X ti ∧ T i
Xtn,T = D0 Xtf ,T = D0
T T
ti ≤t ti ≤t
Inserting Xtn,T , Xtf ,T rather than Xtn , Xtf into formula (4) implies a stock
process S given as
Pt St = (S0 − Xtn,T )Mt − Xtf ,T
where M is a geometric Brownian motion with mean 1 and volatility σ. This

gives rise to affine dividends, namely
(T − ti )+

ti ∧ T
Sti − − Sti = Pt−1 i
D0i Mti +
T T
f ,T
!
+ P S
−1 i (T − ti ) ti ti − + Xti − ti ∧ T
= Pti D0 + = αi + βi Sti −
T S0 − Xtn,T
i−
T
Note that αi & 0 for ti & 0 and βi & 0 for ti % T .

Introduction
Bos-Vandermark
Affine dividends
Conclusion
It is now standard to model dividends as affine functions of the spot at the

ex-date
Dtii = αi + βi Sti − (5)
with for example βi = 0 for ti < 1y and αi = 0 if ti ≥ 5y .
This is equivalent to the stock process being an affine function in the sense that
St = (Ft − Ct )Mt + Ct (6)
holds for some martingale M and a deterministic function C .
This approach dates back to [BFF+ 00] p7f. More recent references include
Overhaus et al. [OFK+ 02] p116ff as well as [OBB+ 07] p6ff.

Introduction
Bos-Vandermark
Affine dividends
Conclusion
More formally, one has (compare Overhaus et al. [OBB+ 07] p7):
Proposition (affine stock model): Assume a stock process S with jumps due
to dividends only. The following two statements are equivalent:
1 The stock process S is positive and pays positive affine dividends (5)
satisfying
αi = 0 for all i ≥ n for some n
βi ∈ [0, 1) for all i
Pti αi ≥ −βi j≥i Ptj αj jk=i (1 − βk )−1 for all i
P Q
2 The stock process S is an affine function (6) of a positive continuous

martingale M with M0 = 1 where C satisfies
Ct = 0 for all t ≥ T for some T
∆Ft ≤ ∆Ct ≤ 0 for all t
dCt = rCt for all t ∈
/ {t1 , t2 , . . . }
with notation ∆ft ≡ ft − ft− for a given càdlàg function f .

Introduction
Bos-Vandermark
Affine dividends
Conclusion

(2) ⇒ (1): Substituting (6) into the dividend drop Sti − − Sti yields
Sti − − Sti = (−∆Fti + ∆Cti )Mti − − ∆Cti
St − − Cti −
= (−∆Fti + ∆Cti ) i − ∆Cti ≥ 0
Fti − − Cti −
and hence
Cti − ∆Fti − Fti − ∆Cti ∆(Fti − Cti )
αi = βi = −
Fti − − Cti − Fti − − Cti −
One computes
Pti+1
αi + βi Cti − = −∆Cti = Cti − − Cti+1 −
Pti
and hence
j
Pti αi Pt C t − X Y
Pti Cti − = + i+1 i+1 = ... = Ptj αj (1 − βk )−1
1 − βi 1 − βi
j≥i k=i
The properties of C imply the required properties of αi and βi for all i as well
as positivity of S.
Introduction
Bos-Vandermark
Affine dividends
Conclusion

(1) ⇒ (2): Both C and M can be constructed by induction on [ti−1 , ti ) if given
on t ≥ ti such that (6) holds:
For t ≥ tn define Ct = 0 as well as Mt through St = Ft Mt .
If C is defined for t ≥ ti set
αi + βi Cti
− ∆Cti = αi + βi Cti − = (7)
1 − βi
This defines C and hence M through (6) on t ≥ ti−1 , and one has
Sti − − Cti −
∆Sti = (∆Fti − ∆Cti ) + ∆Cti = . . . = −βi Sti − − αi
Fti − − Cti −
where we used
αi + βi Fti
−∆Fti = αi + βi Fti − =
1 − βi
Note that as seen above (7) yields via induction
j
X Y
Pti Cti − = Ptj αj (1 − βk )−1
j≥i k=i
Hence C has the required properties. Continuity of M at ti can also be

established.
Introduction
Bos-Vandermark
Affine dividends
Conclusion
Examples
Special cases include

Proportional dividends: Ct = 0
Deterministic dividends up to some final date T :
X
Pt Ct = D0i
T ≥ti >t
Without T one has the deterministic case St = Ft due to (3).

Blend model:
X
Pt Ct = (1 − τi )D0i (8)
ti >t
with some increasing sequence τ in [0, 1].

We have seen above that Bos-Vandermark is a model with affine
dividends. Since Ct ≤ 0 the stock process can be negative.

Introduction
Bos-Vandermark
Affine dividends
Conclusion
So far we only referred to expected discounted dividend D0i as well as dividend

drop Dtii . The next step is to define Dti for t < ti .
We obtain for the affine model (6) from (2)

Dtii = Pt−1
i
D0i − ∆Cti Mti + ∆Cti
For t ≤ ti we may replace Dtii , Pti , Mti by Dti , Pt , Mt , namely

Pt Dti = D0i − Pti ∆Cti Mt + Pti ∆Cti
Summation over all i with ti > t yields

 
X X
Pt St = Pt Ft − Pti ∆Cti  Mt + Pti ∆Cti
ti >t ti >t
= Pt (Ft − Ct )Mt + Pt Ct
We summarize:
Introduction
Bos-Vandermark
Affine dividends
Conclusion
Proposition (affine dividends): Dividend dynamics

dDti = rt Dti dt + Pt−1 D0i − Pti ∆Cti dMt
= rt Dti dt + Pt−1 D0i τi dMt (9)
together with (1) implies the affine stock model (6).
Remarks:
1 Dividend volatility is homogenous in t in the sense that volatility does not
decrease as t approaches ti (assuming M has constant volatility).
2 Dividend dynamics is not homogenous in i since dividend volatility scales
with τi . The effect is non-homogenous equity returns: near returns
experience deterministic dividends while far returns have proportional
dividends.

Introduction Indicator function
Affine dividends Korn-Rogers
Homogenous dividend dynamics Piecewise linear
Conclusion Exponential
Definition: Homogenous dividend dynamics assumes an increasing function τ

with values in [0, 1] and τ0 = 0 as well as
dDti = rt Dti dt + Pt−1 D0i τti −t dMt (10)
The difference with the blend model (9) is that the impact dMt is scaled down
as the ex-dividend date approaches.
The solution is given as

Z t
Pt Dti = D0i 1 + τti −s dMs
0
In case of continuous τ integration by parts yields

Z t
i i
Pt Dt = D0 1 + τti −t Mt − τti M0 − Ms dτti −s
0
Note that M ≥ 0 and M0 = 1 guarantees positive dividends.

Proposition (homogenous dividends): With homogenous dividend dynamics

the stock process is given as sum of an affine function and an integral of the
martingale M. More specifically:
Z t
St = Ft + Pt−1 Ttt Mt − T0t M0 − Ms dTst
0
with Tst for 0 ≤ s ≤ t defined as

X
Tst = D0i τti −s
ti >t
Proof: One has

X
P t St = Pt Dti
ti >t
X Z t
= D0i 1 + τti −t Mt − τti M0 − Ms dτti −s
ti >t 0

Examples: indicator function
Dividend announcement at a fixed time interval a before the ex-date is

represented by the indicator function
τt = 1{t>a}
Assuming M0 = 1 one has

Z t
Pt Dti = D0i 1 + 1{ti −s>a} dMs
0
= D0i Mt∧(ti −a)+
and
 
X X X
St = Dti = Pt−1  D0i Mt + D0i M(ti −a)+  (11)
ti >t ti −a>t ti >t≥ti −a

Examples: Korn-Rogers
The situation discussed by Korn and Rogers [KR05] is obtained by further

specializing
ti = ih, a = (1 − ε)h ≥ 0, Mt = e −µt Xt /X0 , Pt = e −rt
where X is an exponential Lévy process with mean
IE [ Xt ] = X0 e µt
as well as
D0i = λX0 Pti −a e µ(ti −a)

Examples: Korn-Rogers
Formula (11) yields

X
St = λ e (µ−r )((i−1+ε)h−t) Xt
(i−1+ε)h>t
X
+λ e r (t−(i−1+ε)h) X(i−1+ε)h
ih>t≥(i−1+ε)h
The second sum is either empty if the next dividend after t has not been
announced yet or has exactly one term.
Note that formula (11) also covers the case of longer announcement periods,
namely a > h or ε < 0. In that case the second sum consisting of all dividends
announced at t may have more than one term.

Examples: piecewise linear
An alternative to the indicator function is the piecewise linear function

t
τt = ∧1
a
where a is the time distance beyond which a given dividend has maximal
volatility. One obtains for t ≥ a
 
Z t
X i t i − t  Mt + a−1
X
Pt St = Pa+t Fa+t + D0 D0i Ms ds
a (ti −a)+
a+t≥ti >t a+t≥ti >t
For t < a one has to add the term

X i ti
D0 1 −
a
a≥ti >t
representing the announced part of near dividends.

Examples: exponential
With
τt = 1 − e −λt
one obtains

Tst = Pt Ft − e λs Ct
dTst = λPt Ct e λs ds
where Ct is defined through
X
Pt Ct = D0i e −λti
ti >t
Thus
Z t
St = Ct + Ft − e λt Ct Mt + λCt e λs Ms ds
0
The corresponding non-homogenous affine dividend model (6) is

St = Ct + (Ft − Ct ) Mt

Examples: exponential
λ = 0.5, σ = 20%, r = 5%, annual dividends d = 10% paid before maturity

Introduction
Affine dividends Summary
Homogenous dividend dynamics References
Conclusion
Conclusion
Affine stock models (including Bos-Vandermark) can be viewed as dividend

discount models where all dividends are driven by the same martingale
factor. The volatility level of each dividend does not change through time.
Homogenous dividend models behave like bonds in the sense that their
volatility level decreases through time, reaching zero at the ex-date.
The stock return distribution of homogenous dividend models is more
reasonable as one does not shift from affine for near returns to
proportional for far returns.
Modelling homogenous dividend models is more involved since the stock
process is a time average of a postive martingale.

Introduction
Conclusion
Bibliography I
H. Bühler, A. Dhouibi, and D. Sluys.

Stochastic proportional dividends.
http://ssrn.com/abstract=1706758, 2010.
O. Brockhaus, M. Farkas, A. Ferraris, D. Long, and M. Overhaus.

Equity Derivatives and Market Risk Models.
Risk Books, February 2000.
G. Bernhart and J. Mai.

Consistent modeling of discrete cash dividends.
Technical report, Xaia Investment, 2012.
M. Bos and S. Vandermark.

Finessing fixed dividends.
Risk, pages 157–158, 2002.
R. Korn and L. C. G. Rogers.

Stocks paying discrete dividends modeling and option pricing.
Journal of Derivatives, 13(2):44–48, 2005.
M. Overhaus, A. Bermúdez, H. Bühler, A. Ferraris, C. Jordinson, and A. Lamnouar.

Equity Hybrid Derivatives.
John Wiley and Sons, 2007.
M. Overhaus, A. Ferraris, T. Knudsen, R. Milward, L. Nguyen-Ngoc, and G. Schindlmayr.

Equity Derivatives: Theory and Applications.
John Wiley and Sons, February 2002.

Introduction
Conclusion
Bibliography II
R. Tunaru.
Dividend derivatives.
ssrn, 2014.
J. van Binsbergen, M. Brandt, and R Koijen.

On the timing and pricing of dividends.
Technical report, National Bureau of Economic Research, 2010.
M.H. Vellekoop and M.H. Nieuwenhuis.

Efficient pricing of derivatives on assets with discrete dividends.
Applied Mathematical Finance, 13(3):265–284, 2006.

Appendix: stochastic dividend yield
Bühler [BDS10] suggests an Ornstein-Uhlenbeck process y driving dividend

yield. The resulting dynamics can be written as
St = F t Nt i Mt ti ≤ t < ti+1
where N is an adapted process of the form

S0 − Pt ≤t di S0 − Pt ≤t Ci +Di +Ei yt P
− t ≤t Ci +Ei yt
Nt = e i = e i = e i
Ft Ft
with IE [ Mti Nti ] = 1, Pti Fti = e −Di Pti−1 Fti−1 as well as Ei ∈ {1, Di }.
Remark: It is not required that N or NM is a martingale and indeed

h i
IEti−1 [ Nti ] = Nti−1 e −Ci IEti−1 e −Ei yti 6= Nti−1

Appendix: stochastic dividend yield
The dividend drop Dtii satisfies
Pti Dtii

= −Pti ∆Sti = Pti−1 Fti−1 Nti−1 − Pti Fti Nti Mti
A consistent dividend dynamics for t ≤ ti can be defined as
Pt Dti =

Pti−1 Fti−1 Nti−1 ∧t − Pti Fti Nti ∧t Mt
For t ≤ ti−1 this simplifies to
Pt Dti = D0i Nt Mt
One observes
X
Pt St = Pt Dti = Pti−1 Fti−1 Nti−1 Mt ti−1 ≤ t < ti
ti >t
as required. Bühler [BDS10] also discusses affine dividends

Nt i
Dtii = αi + βi Sti − = αi + 1 − Sti −
Nti−1
Analysis of a version of this model with homogenous dividend dynamics is left
for future research.

Brockhaus Oliver 03.12.2014

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Brockhaus Oliver 03.12.2014

Hochgeladen von

Copyright:

Verfügbare Formate

Introduction

Dividend Discount Approach to Equity Derivatives Modelling

Cass Business School

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 1 / 32

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 2 / 32

There are various directions of research on the topic of dividends:

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 4 / 32

Discrete dividends with known future ex-dividend dates

with deterministic fi . In practice one often assumes

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 5 / 32

Dividend discount models

Sti − − Sti = Dtii (2)

In this presentation we address the following questions:

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 6 / 32

D0i = D0k e −q∆(i−k) and ti = tk + ∆(i − k)

for all i ≥ k and some time increment ∆ one computes

from (1) or (3) with t = 0. Growth parameter q can be interpreted as long

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 7 / 32

with deterministic functions

This formula is a good approximation for a European option price with

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 9 / 32

We now fix t independently of the product, namely

Pt St = (S0 − Xtn,T )Mt − Xtf ,T

where M is a geometric Brownian motion with mean 1 and volatility σ. This

Note that αi & 0 for ti & 0 and βi & 0 for ti % T .

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 10 / 32

Affine stock process

It is now standard to model dividends as affine functions of the spot at the

Dtii = αi + βi Sti − (5)

with for example βi = 0 for ti < 1y and αi = 0 if ti ≥ 5y .

St = (Ft − Ct )Mt + Ct (6)

holds for some martingale M and a deterministic function C .

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 11 / 32

Affine stock process

2 The stock process S is an affine function (6) of a positive continuous

with notation ∆ft ≡ ft − ft− for a given càdlàg function f .

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 12 / 32

Affine stock process

Affine stock process

Hence C has the required properties. Continuity of M at ti can also be

Special cases include

Without T one has the deterministic case St = Ft due to (3).

with some increasing sequence τ in [0, 1].

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 15 / 32

Affine dividend dynamics

So far we only referred to expected discounted dividend D0i as well as dividend

We obtain for the affine model (6) from (2)

For t ≤ ti we may replace Dtii , Pti , Mti by Dti , Pt , Mt , namely

Summation over all i with ti > t yields

Affine dividend dynamics

Proposition (affine dividends): Dividend dynamics

= rt Dti dt + Pt−1 D0i τi dMt (9)

together with (1) implies the affine stock model (6).

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 17 / 32

Homogenous dividend dynamics

Definition: Homogenous dividend dynamics assumes an increasing function τ

dDti = rt Dti dt + Pt−1 D0i τti −t dMt (10)

The solution is given as

In case of continuous τ integration by parts yields

Note that M ≥ 0 and M0 = 1 guarantees positive dividends.

oliver.brockhaus@mathfinance.com Dividend Discount Approach to Equity Derivatives Modelling 19 / 32

Homogenous dividend dynamics

Proposition (homogenous dividends): With homogenous dividend dynamics

with Tst for 0 ≤ s ≤ t defined as