Negative Probabilities in Financial Modeling

Gunter Meissner
Mark Burgin
Shidler College of Business
Department of Mathematics
University of Hawaii
University of California, Los Angeles
2404 Maile Way, D202
405 Hilgard Ave.
Honolulu, HI 96822
Los Angeles, CA 90095
Email: [email protected]

Abstract

We first define and derive general properties of negative probabilities. We then show how
negative probabilities can be applied to modeling financial options such as Caps, Floors and
Swapoptions. In trading practice, these options are typically valued in a Black-Scholes-Merton
framework assuming a log-normal distribution for the underlying interest rate. However, in some
cases, such as the 2008/2009 financial crisis, interest rates can get negative. Then the log-normal
distribution is inapplicable. We show how negative probabilities can be applied to value interest
rate options in a log-normal framework implying a positive probability for negative interest rates.
A model in VBA, which prices Caps, Floors and Swapoptions with negative probabilities, is
available upon request.

Key words: Negative probabilities, negative interest rates, Caps, Floors, Swaptions

JEL Classification: C10

[Type text]
1. Introduction
The classical probability theory is applied in most sciences and in many of the humanities. In
particular, it is successfully used in physics and finance. However, physicists found that they
need a more general approach than the classical probability theory. The first was Eugene Wigner
(1932), who introduced a function, which looked like a conventional probability distribution and
has later been better known as the Wigner quasi-probability distribution because in contrast to
conventional probability distributions, it took negative values, which could not be eliminated or
made nonnegative. The importance of Wigner's discovery for foundational problems was not
recognized until much later. Another outstanding physicist, Dirac (1942) not only supported
Wigner’s approach but also introduced the physical concept of negative energy. He wrote:

“Negative energies and probabilities should not be considered as nonsense. They are
well-defined concepts mathematically, like a negative of money."
After this, negative probabilities little by little have become popular although questionable
techniques in physics. Bartlett (1945) worked out the mathematical and logical consistency of
negative probabilities. However, he did not establish rigorous foundation for negative
probability utilization. Khrennikov (2009) provides the first mathematical theory of negative
probabilities in his textbook. However, he is doing this not in the conventional setting of real
numbers but in the framework of p-adic analysis.
Negative probabilities are also used in mathematical finance. The concept of risk-neutral or
pseudo probabilities is a popular concept and has been numerously applied, for example, in
credit modeling by Jarrow and Turnbull (1995), and Duffie and Singleton (1999). Haug (2007)
extends the risk-neutral framework to allow negative probabilities and shows how negative
probabilities can help add flexibility to financial modeling.
The remaining paper is organized as follows. In section 2, we resolve the mathematical issue
of the negative probability problem. We build a mathematical theory of extended probability as a
probability function, which is defined for real numbers and can take both positive and negative
values. Thus, extended probabilities include negative probabilities. Different properties of
extended probabilities are found. In section 3, we give examples of negative nominal interest
rates in financial practice and show problems of current financial modeling of negative interest
rates. In Section 4, we build mathematical models of interest rate options as Caps and Floors and
Swapoptions, integrating extended probabilities into the pricing model to allow for negative

[Type text]
interest rates. Conclusions are given in Section 5. A follow up paper will specify >1 probabilities
and apply them to financial options.

2. Mathematical theory of extended probability

Extended probabilities generalize the standard definition of a probability function. At first, we

define extended probabilities in an axiomatic way and then develop application of extended
probabilities to finance.
To define extended probability, EP, we need some concepts and constructions, which are
described below. Some of them are well-known, such as, for example, set algebra, while others,
such as, for example, random antievents, are new.
We remind that if X is a set, then | X| is the number of elements in (cardinality of) X
(Kuratowski and Mostowski, 1967). If A ⊆ X, then the complement of A in X is defined as CXA =
X \ A.
A system B of sets is called a set ring (Kolmogorov and Fomin, 1989) if it satisfies
conditions (R1) and (R2):
(R1) A, B ∈ B implies A ∩ B ∈ B.
(R2) A, B ∈ B implies A ∆ B ∈ B where A ∆ B = (A \ B) ∪ (B \ A).
For any set ring B, we have ∅ ∈ B and A, B ∈ B implies A ∪ B, A \ B ∈ B.
Indeed, if A ∈ B, then by R1, A \ A = ∅ ∈ B. If A, B ∈ B, then A \ B = ((A \ B) ∪ (B \ A))
∩ A ∈ B. If A, B ∈ B and A ∩ B = Ø, then A ∆ B = A ∪ B ∈ B. It implies that A ∪ B = (A \ B)
∪ (B \ A) ∪ (A ∩ B) ∈ B. Thus, a system B of sets is a set ring if and only if it is closed with
respect union, intersection and set difference.
Example 2.1. The set CI of all closed intervals [a, b] in the real line R is a set ring.
Example 2.2. The set OI of all open intervals (a, b) in the real line R is a set ring.
A set ring B with a unit element, i.e., an element E from B such that for any A from B , we
have A ∩ E = A, is called a set algebra (Kolmogorov and Fomin, 1989).
Example 2.3. The set BCI of all closed subintervals of the interval [a, b] is a set algebra.
Example 2.4. The set BOI of all open subintervals of the interval [a, b] is a set algebra.

[Type text]
 A set algebra B closed with respect to complement is called a set field.
Let us consider a set Ω, which consists of two irreducible parts (subsets) Ω+ and Ω-, i.e.,
neither of these parts is equal to its proper subset, a set F of subsets of Ω, and a function P from
F to the set R of real numbers.
Elements from F, i.e., subsets of Ω that belong to F, are called random events.
Elements from F + = {X ∈ F; X ⊆ Ω+ } are called positive random events.
Elements from Ω+ that belong to F+ are called elementary positive random events or simply,
elementary positive random events.
If w ∈ Ω+, then –w is called the antievent of w.
Elements from Ω- that belong to F- are called elementary negative random events or
elementary random antievents.
For any set X ⊆ Ω+, we define
X + = X ∩ Ω+,
X - = X ∩ Ω -,
–X = { -w; w ∈ X}
and
F - = { –A ; A ∈ F+ }
If A ∈ F+ , then –A is called the antievent of A.
Elements from F - are called negative random events or random antievents.
Definition 1. The function P from F to the set R of real numbers is called a probability
function, if it satisfies the following axioms:
EP 1 (Order structure). There is a graded involution α: Ω → Ω, i.e., a mapping such that α2
is an identity mapping on Ω with the following properties: α(w) = -w for any element w from Ω,
α(Ω+) ⊇ Ω-, and if w ∈ Ω+, then α(w) ∉ Ω+.
EP 2 (Algebraic structure). F + ≡ {X ∈ F; X ⊆ Ω+ } is a set algebra that has Ω+ as a member.
EP 3 (Normalization). P(Ω+) = 1.
EP 4 (Composition) F ≡{X; X+⊆ F+ & X -⊆ F- & X + ∩ -X – ≡ ∅ & X - ∩ -X + ≡ ∅}.
EP 5 (Finite additivity)
P(A ∪ B) = P(A) + P(B)
for all sets A, B ∈ F such that

[Type text]
 A∩B≡Ø
EP 6 (Annihilation). { vi , w, - w ; vi , w ∈ Ω & i ∈ I } = { vi; vi ∈ Ω & i ∈ I } for any element
w from Ω.
Axiom EP6 shows that if w and - w are taken (come) into one set, they annihilate one
another. Having this in mind, we use two equality symbols = and ≡. The second symbol means
equality of elements of sets. The second symbol also means equality of sets, when two sets are
equal when they have exactly the same elements (Kuratowski and Mostowski, 1967). The
equality symbol = is used to denote equality of two sets with annihilation, for example, { w, - w }
= Ø. Note that for sets, equality ≡ implies equality =.
For equality of numbers, we, as it is customary, use symbol =.
EP 7. (Adequacy) A = B implies P(A) = P(B) for all sets A, B ∈ F.
For instance, P({ w, - w }) = P(Ø) = 0.
EP 8. (Non-negativity) P(A) ≥ 0, for all A ∈ F+.
It is known that for any set algebra A, the empty set Ø belongs to A and for any set field B in
Ω, the set Ω belongs to A (Kolmogorov and Fomin, 1989).
Definition 2. The triad (Ω, F, P) is called an extended probability space.
Definition 3. If A ∈ F, then the number P(A) is called the extended probability of the event
A.
Let us obtain some properties of the introduced constructions.
Lemma 1. α(Ω+) ≡ -Ω+ ≡ Ω- and α(Ω-) ≡ -Ω- ≡ Ω+.
Proof. By Axiom EP1, α(Ω+) ≡ -Ω+ and α(Ω+) ⊇ Ω-. As Ω ≡ Ω+ ∪ Ω- , Axiom EP1 also
implies α(Ω+) ⊆ Ω-. Thus, we have α(Ω+) ≡ Ω-. The first part is proved.
The second part is proved in a similar way.
Thus, if Ω+ = { wi ; i ∈ I }, then Ω- = { -wi ; i ∈ I }.
As α is an involution of the whole space, we have the following result.
Proposition 1. α is a one-to-one mapping and | Ω+ | = |Ω-|.
Corollary 1. (Domain symmetry) w ∈ Ω+ if and only if –w ∈ Ω-.
Corollary 2. (Element symmetry) - (- w ) = w for any element w from Ω.
Corollary 3. (Event symmetry) - (- X ) ≡ X for any event X from Ω.
Lemma 2. α(w) ≠ w for any element w from Ω.

[Type text]
 Indeed, this is true because if w ∈ Ω+, then by Axiom EP1, α(w) ∉ Ω+ and thus, α(w) ≠ w. If
w ∈ Ω-, then we may assume that α(w) = w. However, in this case, α(v) = w for some element v
from Ω+ because by Axiom EP1, α is a projection of Ω+ onto Ω-. Consequently, we have
α(α(v)) = α(w) = w
However, α is an involution, and we have α(α(v)) = v. This results in the equality
v=w
Consequently, we have α(v) = v. This contradicts Axiom EP1 because v ∈ Ω+. Thus, lemma
is proved by contradiction.
Proposition 2. Ω+ ∩ Ω- ≡ Ø.
Proposition 3. F+ ⊆ F , F - ⊆ F and F ⊆ F+ ∪ F -.
Corollary 1 implies the following result.
Proposition 4. X ⊆ Ω+ if and only if - X ⊆ Ω-.
Proposition 5. F - ≡ {X ∈ F; X ⊆ Ω- } = F ∩ Ω-.
Corollary 4. F+ ∩ F- ≡ Ø.
Axioms EP6 implies the following result.
Lemma 3. X ∪ -X = Ø for any subset X of Ω.
Indeed, for any w from the set X, there is -w in the set X, which annihilates w.
Let us define the union with annihilation of two subsets X and Y of Ω by the following
formula:
X + Y ≡ (X ∪ Y) \ [(X ∩ -Y) ∪ (-X ∩ Y)]
Here the set-theoretical operation \ represents annihilation, while sets X ∩ -Y and X ∩ -Y
depict annihilating entities.
Some properties of the new set operation + are the same as properties of the union ∪, while
other properties are different. For instance, there is no distributivity between operations + and ∩.
Lemma 4. a) X + X ≡ X for any subset X of Ω;
b) X + Y ≡ X + Y for any subsets X and Y of Ω;
c) X + Ø ≡ X for any subset X of Ω;
d) X + (Y + Z) ≡ (X + Y) + Z for any subsets X , Y and Z of Ω;
e) X + Y ≡ X ∪ Y for any subsets X and Y of Ω+ (of Ω-);
Lemma 5. a) Z ∩ (X + Y ) ≠ Z ∩ X + Z ∩ Y ;

[Type text]
 b) X + (Y ∩ Z) ≠ (X ∩ Y) + (X ∩ Z).
Lemma 6. A ∩ B ≡ (A+ ∩ B+) + (A- ∩ B-) for any subsets A and B of Ω.
Indeed, as A ≡ A+ ∪ A- and B ≡ B+ ∪ B-, we have
A ∩ B ≡ (A+ ∪ A- ) ∩ (B+ ∩ B-) ≡
(A+ ∩ B+) ∪ (A+ ∩ B-) ∪ (A- ∩ B+) ∪ (A- ∩ B-) ≡
(A+ ∩ B+) + (A- ∩ B-)
because (A+ ∩ B-) ≡ Ø and (A- ∩ B+) ≡ Ø.
In a similar way, we prove the following results.
Lemma 7. A \ B ≡ (A+ \ B+) + (A- \ B-) for any subsets A and B of Ω.
Lemma 8. X ≡ X+ + X- = X + ∪ X - for any set X from F.
Lemma 9. A + B ≡ (A+ + B+) + (A-+ B-) for any sets X and Y from F.
Axioms EP6 and EP7 imply the following result.
Proposition 6. P(X + Y ) = P(X ∪Y) for any two events X and Y from Ω.
Lemma 10. P(∅) = 0.
Properties of the structure F + are inherited by the structure F .
Theorem 1. (Algebra symmetry) If F + is a set algebra (or set field), then F is a set field (or
set algebra) with respect to operations + and ∩.
Proof. At first, we prove that F- is a set algebra (or set field).
Let us assume that F + is a set algebra and take two negative random events H and K from F -.
By the definition of F -, H = -A and K = -B for some positive random events A and B from F+.
Then we have
H ∩ K = (- A) ∩ (-B) = -( A ∩ B)
As F + is a set algebra, A ∩ B ∈ F+. Thus, H ∩ K ∈ F-.
In a similar way, we have
H ∪ K = (- A) ∪ (-B) = -( A ∪ B)
As F + is a set algebra, A ∪ B ∈ F+. Thus, H ∪ K ∈ F-.
By the same token, we have H \ K ∈ F-.
Besides, if F+ has a unit element E, then –E is a unit element in F-.
Thus, F - is a set algebra.

[Type text]
 Now let us assume that F + is a set field and H ∈ F -. Then by the definition of F -, H = -A for
a positive random event A from F+. It means that CΩ+A = Ω+ \ A ∈ F+. At the same time,
CΩ-H = Ω- \ H = (-Ω+ ) \ (-A) = -(Ω+ \ A) = -CΩ+A
As CΩ+A belongs to F+, the complement CΩ-H of H belongs to F-. Consequently, F - is a set
field.
Let us once more assume that F+ is a set algebra and take two random events A and B from F.
Then by Theorem 1, F- is a set algebra. By Lemma 8, A ≡ A+ + A- and B ≡ B+ + B-. By Axiom
EP4, A+, B+ ∈ F+, A-, B- ∈ F-, while by Proposition 2, A + ∩ A – ≡ ∅, B + ∩ B – ≡ ∅, A ≡ A+ ∪ A-,
and B ≡ B+ ∪ B-.
By Lemma 6, A ∩ B ≡ (A+ ∩ B+) + (A- ∩ B-). Thus, (A ∩ B)+ ≡ A+ ∩ B+ and (A ∩ B)- ≡ A- ∩
B-. As F+ is a set algebra, (A ∩ B)+ ≡ A+ ∩ B+∈ F+. As it is proved that F- is a set algebra, (A ∩
B)- ≡ A- ∩ B-∈ F-. Consequently, A ∩ B ∈ F.
By Lemma 7, A \ B ≡ (A+ \ B+) + (A- \ B-). Thus, (A \ B)+ ≡ A+ \ B+ and (A \ B)- ≡ A- \ B-. As F+
is a set algebra, (A \ B)+ ≡ A+ \ B+∈ F+. As it is proved that F- is a set algebra, (A \ B)- ≡ A- \ B-∈
F-. Consequently, A \ B ∈ F.
By Lemma 9, A + B ≡ (A+ + B+) + (A- + B-). Thus, (A + B)+ ≡ A+ + B+ and (A + B)- ≡ A- + B-.
As F+ is a set algebra, (A + B)+ ≡ A+ + B+ ≡ A+ ∪ B+∈ F+. As it is proved that F- is a set algebra,
(A + B)- ≡ A- + B- ≡ A- ∪ B- ∈ F-. Consequently, A + B ∈ F.
Besides, if F+ has a unit element E, then –E is a unit element in F- and E ∪ -E is a unit
element in F.
Thus, F is a set algebra.
Now let us assume that F + is a set field and A ∈ F. Then as it is demonstrated above, F - is a
set field. By Lemma 8, A ≡ A+ + A-. By Proposition 2, Ω+ ∩ Ω- = Ø, we have
CΩA = CΩ+A + CΩ-A
Then CΩ+A belongs to F+ as F+ is a set field and as it is proved in Theorem 1, CΩ-A belongs
to F-. Consequently, CΩA belongs to F- and F - is a set field.
Theorem is proved.

[Type text]
 3. Negative interest rates and the problem of their modeling

Negative probabilities can help to model interest rates and interest rate derivatives. To show this,
let us start with the equation

Real interest rate = Nominal interest rate – Inflation rate (1)

where

the nominal interest rate is the de facto rate, which is received by the lender and paid by the
borrower in a financial contract. For example, the nominal interest rate is the rate, which the
lender receives on a saving account or the coupon of a bond. From equation (1) we see that a real
interest rate can easily be negative and in reality often is. For example, if the nominal interest
rate on a savings account is 1% and the inflation rate is 3%, naturally, the real interest rate, i.e.
the inflation adjusted rate of return for the lender is -2%.

3.1. Examples of negative nominal interest rates

However, in rare cases, also the nominal interest rate can be negative. An example of this would
be that the lender gives money to a bank, and additionally gives pays the bank an interest rate.
This happened in the 1970s in Switzerland. The lender had several motives

a) Switzerland is considered an extremely safe country to place capital

b) Investors were speculating on an increase of the Swiss franc
c) Some investors avoided paying taxes in their home country

Another example of negative nominal interest rates occurred in Japan in 2003. Banks lent
Japanese Yen and were willing to receive a lower Yen amount back several days later. This
means de facto a negative nominal interest rate. The reason for this unusual practice was that
banks were eager to reduce their exposure to Japanese Yen, since confidence in the Japanese
economy was low and the Yen was assumed to devalue.

[Type text]
Similarly, in the US, from August to November 2003, ‘repos’, i.e. repurchase agreements traded
at negative interest rates. A repo is just a collateralized loan, i.e. the borrower of money gives
collateral, for example a Treasury bond, to the lender for the time of the loan. When the loan is
paid back, the lender returns the collateral. However, in 2003 in the US, settlement problems
when returning the collateral occurred. Hence the borrower was only willing to take the risk of
not having the collateral returned if he could pay back a lower amount than originally borrowed.
This constituted a negative nominal interest rate.

A further example of the market expecting the possibility of negative nominal interest rates
occurred in the worldwide 2008/2009 financial crisis, when strikes on options on Eurodollars
Futures contracts were quoted above 100. A Eurodollar is a dollar invested at commercial banks
outside the US. A Eurodollar futures price reflects the anticipated future interest rate. The rate is
calculated by subtracting the Futures price from 100. For example, if the 3 month March
Eurodollar future price is 98.5, the expected interest rate from March to June is 100 – 98.5 = 1.5,
which is quoted in per cent, so 1.5%. In March 2009, option strikes on Eurodollar future
contracts were quoted above 100 on the CME, Chicago Mercantile Exchange. This means that
market participants could buy the right to pay a negative nominal interest on US dollars in the
future if desired. The reason for this unusual behavior is that investors wanted to invest in the
safe haven currency US dollar even if they had to pay for it.

3.2. Modeling interest rates

In finance, interest rates are typically modeled with a geometric Brownian motion,

dr
= μ r dt + σ r ε dt (2)
r

dr : change in the interest rate r;

µ r : drift rate, which is the expected growth rate of r, assumed non-stochastic and constant

dt : infinitely short time period

σ r : expected volatility of rate r, assumed non-stochastic and constant

[Type text]
ε : random drawing from a standardized normal distribution. All drawings at times t are
independent from each other.

In equation (2), the first term on the right hand side gives the average growth rate of r. The
second term on the ride side adds stochasticity to the process via ε, i.e. provides the distribution
around the average growth rate. Importantly, from equation (1) we can observe that the relative
change dr/r is normally distributed, since ε is normally distributed. If the relative change of a
variable is normally distributed, it follows that the variable itself is log-normally distributed with
a pdf

1  ln(x) −μ  2
1 -  
2 σ 
e (3)
x σ 2π

In the equation (3), μ and σ are the mean and standard deviation of ln(x) respectively. Figure 1
shows a log-normal distribution.

0.6

0.5

0.4

0.3

0.2

0.1

0
0.01 0.51 1.01 1.51 2.01 2.51 3.01

Figure 1 Log-normal distribution with μ = 0 and σ =1.

The logarithm of a negative number is not defined, hence with the pdf equation (3), negative
values of interest rates cannot be modeled. However, as discussed above, negative interest rates
do exist in the real financial world. Here negative probabilities come into play. We will explain
this with options on interest rates.

[Type text]
 4. How negative probabilities allow more adequate interest rate modeling

4.1. Modeling interest rate options

Two main types of options are call options and put options.

A call option reflects the right but not the obligation to pay a strike price and receive an
underlying asset.

A put option reflects the right but not the obligation to receive a strike price and deliver an
underlying asset.

Let’s derive the equations used in financial practice to value calls and puts. From equation (2),
applying Ito’s lemma, we derive the famous 1997 Nobel Prize rewarded work of Black, Scholes
and Merton, which resulted in the PDE 1

∂D 1 ∂D 1 ∂ 2D 1 2 2
D= + S+ σS (4)
∂t i ∂S 2 ∂S2 i

where

D : financial derivatives as for example a call option or a put option

i: discount rate

S : modeled variable

σ : volatility of S

One equation that satisfies the PDE (4) is the Black-Scholes-Merton equation for a call and a
put 2. For a call option, we have

1
For a proof see www.dersoft.com/BSMPDEgeneration.pptx
2
For a proof, see www.dersoft.com/bspdeproof.doc

[Type text]
 S0 1
ln( −iT
) + σ 2T
C = S0 N(d1 ) − Ke − iT N(d 2 ) with d1 = Ke 2 and d 2 = d1 − σ T (5)
σ T

where

N(x) : cumulative probability of a standard normal distribution at x

K : Strike, i.e. the price that the buyer may pay at option maturity T to receive the underlying
asset

4.2. Caps and Floors

For interest rate options, a term structure of interest rates exists (i.e. market given interest
rates for different maturities exist). This property is utilized when valuing interest rate options in
the model Black 1976 model. We will first discuss an interest rate option contract on short term
interest rates, i.e. Caps and Floors. A Cap consists of several Caplets. A Caplet is the option but
not the obligation to pay an interest rate rK at option maturity tx. A Floor consists of several
Floorlets. A Floorlet is the option but not the obligation to receive an interest rate rK at option
maturity tx.

A Cap is typically used as an insurance against rising interest rates: If a borrower pays a floating
interest rate on his loan, he can protect himself against rising interest rates by buying a Cap.
Conversely, a Floor is typically used as an insurance for decreasing interest rates. If an investor
receives a floating interest rate on a bond, the investor can protect himself against decreasing
interest rate by buying a Floor.

Using forward interest rates rf, equation (5) becomes

r 1
ln( f ) + σ 2 t x
Caplet ts,tl = m e-rl tl {rf N(d1) – rk N(d2)} with d1 = rk 2 and d2 = d1 - σ t l
σ tx

(6)

[Type text]
Floorlet ts,tl = m PA e-rl tl {rk N(-d2) – rf N(-d1)} (7)

where d1 and d2 are defined as in (6)

Caplet : option on an interest rate from time ts to time tl, tl > ts, i.e. the right but not the obligation
to pay the rate rK at tl.

Floorlet : option on an interest rate from time ts to time tl, tl > ts, i.e. the right but not the
obligation to receive the rate rK at option maturity tl.

m : time between tl and ts, expressed in years

tx: option maturity, tx ≤ ts < tl

 df   1 
rf : forward interest rate, derived as rf t s ,t l = 
ts
− 1   where df is a discount factor, i.e.
 tl   tl − ts
df 
dfty = 1/(1+ry).

rK : strike rate i.e. the interest rate that the Caplet buyer may pay and the Floorlet buyer may
receive at option maturity tx.

For more details, see Meissner 1998.

4.3. Applying negative probabilities to Caplets and Floorlets

Our original problem is that the market applied log-normal distribution, which is underlying the
valuation of interest rate derivatives, cannot model negative interest rates. Several solutions to
this problem are possible.

a) We can model interest rates with an entirely different distribution as for example the
normal distribution, which allows negative interest rates. This is done by Vasicek (1977),
Ho and Lee (1986), and Hull and White (1990). However, empirical data shows that
interest rate distribution behaves far more log-normal than normal. Thus, the suggested
solutions do not correctly reflect the reality.

[Type text]
 b) We can add a location parameter to the log-normal distribution. Hence equation (3)
1  ln(x) −μ  2 1  ln(x -α) −μ  2
1 -   1 -  
σ σ
e 2 
becomes e 2 
, where α is the location
x σ 2π (x - α) σ 2π
parameter. For α > 0, the log-normal distribution is shifted to the left. As a result, the
probability distribution acquires negative values. This is impossible in the conventional
probability theory but fits well into extended probability theory. At the same time,
probability distributions that take negative values are important for practice, allowing to
model negative interest rates.

c) A further way to model options on negative interest rates is to apply negative

probabilities to equations (6) and (7). We add a parameter β to equations (6) and (7)

Caplet t ,t = m e-rl tl {rf [N (d1)-β] – rk [N(d2)-β]}

s l β ∈ ℜ (8)

Floorlet ts,tl = m e-rl tl {rk [N(-d2)-β] – rf [N(-d1)-β]} (9)

This also brings us to negative probabilities. If the strike rK is smaller than the forward
rate sf , for a positive β , negative probabilities may emerge, i.e. N(d1)- β, N(d2)- β, N(-

d2)- β and N(-d1)- β may become negative. Thus, appearance of negative probabilities

depends on the value of β and the option input parameters rK, rF, rl, and σ. The higher the
value of β, the more likely it is that negative probabilities will emerge.

Importantly, applying negative probabilities in equations (8) and (9) decreases the value
of the Caplet and increase the value of the Floorlet. This is an adequate result since it
adjusts the option prices for the possibility of negative interest rates. The magnitude of
the parameter β, that a trader applies, reflects a trader’s opinion on the probability of
negative rates. A trader will use more extreme β-values if he/she believes strongly in the
possibility of negative interest rates, vice versa.

4.4. Swapoptions

[Type text]
Another popular type of interest rate option is a Swapoption. Two types exist, Payers and
Receivers. A Payers Swapoption is the right but not the obligation to pay a fixed interest rate, i.e.
5%, and receive a floating interest rates, e.g. 3-month Libor 3. A Receivers Swapoption is the
right but not the obligation to receive a fixed interest rate, i.e. 5%, and pay a floating interest
rates, e.g. 3-month Libor. Similar to Caps and Floors, Swapoptions are typically used to protect
against interest rate volatility. For example, a company is bidding on an investment project and is
concerned about rising interest rates until the bidding is decided. The company can enter into a
Payers Swapoption. If the company wins the bid and interest rates have risen, the Payers will
allow the company to pay the lower strike rate, which was agreed in the Swapoption.

A Payers Swapoption can be valued by modifying equation (5). This gives us an equation for
Payers Swapoption (PSWO):

n
PA i ( t i − t i −1 ) -rT
PSWO = ∑ (1 + sr )
i =1
( p t i −T ) e ( sr N(d1) - rk N(d2) )
f
(10)
f

where

PAi is the principal amount for the period from ti-1 to ti

srf is the forward swap rate, for the period from ti-1 to tn.

n
srf = (dft - dft ) /
i-1 n ∑ df
i =1
ti
( t i − t i −1 ) and df is a standard discount factor, i.e. dfl = 1/(1+rl).

sr 1
ln( f ) + σ 2 T
rk 2
d1 = and d2 = d1 - σ T
σ T

pt is the payment date of the fixed cash flows

i

Similarly, the Receiver Swapoption (RSWO) can be valued as

3
Libor stands for London Interbank offered rate. It is determined every business day at 11 AM London time.

[Type text]
 n
PA i ( t i − t i −1 ) -rT
RSWO = ∑ (1 + sr )
i =1
( p t i −T )
e ( rk N(-d2) - srf N(-d1) ) (11)
f

For more details see Meissner, 1998.

We once more encounter a problem because the market applied log-normal distribution, which is
underlying the valuation of interest rate derivatives, cannot model negative interest rates. To
solve this problem, we apply the approach used for Caps and Floors to Payers and Receivers
Options.

In this case, it is again possible as in (a), to use an entirely different distribution as the normal
distribution or as in (b), to add a location parameter to the log-normal distribution. A third
possibility is to apply negative probabilities once more. We can add a parameter γ to equations
(10) and (11):

n
PA i ( t i − t i −1 ) -rT
PSWO = ∑ (1 + sr )
i =1
( p t i −T ) e ( sr [N(d1)-γ]- rk [N(d2)-γ] )
f
(12)
f

and

n
PA i ( t i − t i −1 ) -rT
RSWO = ∑ (1 + sr )
i =1
( p t i −T )
e ( rk [N(-d2)- γ] - srf [N(-d1)- γ] ) (13)
f

We derive equivalent results as we did for Caps and Floors. If the strike rK is smaller than the
forward rate srf , for a positive γ, negative probabilities may emerge, i.e. N(d1)-γ, N(d2)-γ, N(-

d2)- γ and N(-d1)- γ may become negative. Thus, appearance of negative probabilities depends

on the value of γ and the option input parameters rK, srf, r, T and σ. The higher the value of γ, the
more likely it is that negative probabilities will emerge.

Importantly, applying negative probabilities decreases the value of the Payers Swapoption and
increases the value of the Receivers Swapoption. This is the desired result since it adjusts the
option prices for the possibility of negative interest rates. The magnitude of parameter γ that a

[Type text]
trader applies, reflects a trader’s opinion on the probability of negative rates. A trader will use
more extreme γ-values if he/she believes strongly in the possibility of negative interest rates, vice
versa.

5. Concluding Summary

We have defined extended probabilities, which include negative probabilities, and

derived their general properties. Then we have applied extended probabilities to financial
modeling. We have shown that negative nominal interest rates have occurred several times in the
past in financial practice, as in the 2008/2009 global financial crisis. This is inconsistent with the
conventional theoretical models of interest rates, which typically apply a log-normal distribution.
In particular, when Caps, Floors and Swapoption are valued in a log-normal Black-Scholes-
Merton framework, then the probability of negative interest rates is zero. Here negative
probabilities come into play. We have shown that integrating negative probabilities in the Black-
Scholes-Merton framework allows to consistently model negative nominal interest rates, which
exist in financial practice.

References:

Bartlett, M. S. (1945). "Negative Probability". Math Proc Camb Phil Soc 41: 71–73
Black F. and M. Scholes, “The Pricing of Options and Corporate Liabilities,” The Journal of
Political Economy, May-June 1973, p.637-659
Black, F., “The Pricing of Commodity Contracts,” Journal of Financial Economics, 3 (March
1976), 167-79.
Dirac, P.A.M. (1942) The Physical Interpretation of Quantum Mechanics, Proc. Roy. Soc.
London, (A 180), pp. 1–39

Duffie D., and K. Singleton, “Modeling Term Structures of Defaultable Bonds”, Review of
Financial Studies, 12 (1999), p.687-720
Ho T., and S. Lee, “Term Structure Movements and Pricing Interest Rate Contingent Claims,”
Journal of Finance, 41, 1986, p.1011-1029

[Type text]
Haug, E. G. (2007): Derivatives Models on Models, John Wiley & Sons, New York
Hull J., and A. White, “Pricing Interest Rate Derivatives Securities,” Review of Financial
Studies, 3, no.4 (1990), p. 573-592

Jarrow R., and S. Turnbull, “Pricing Derivatives on Financial Securities Subject to Credit Risk”,
Journal of Finance, Vol L, No. 1, March 1995, p.53-85

Khrennikov, A. Interpretations of Probability, Walter de Gruyter, Berlin/New York, 2009

Kolmogorov, A. N. (1933) Grundbegriffe der Wahrscheinlichkeitrechnung, Ergebnisse der

Mathematik (English translation: (1950) Foundations of the Theory of Probability, Chelsea P.C.)

A.N. Kolmogorov and S.V. Fomin, Elements of Function Theory and Functional Analysis,
Nauka, Moscow, 1989 (in Russian)

Kuratowski, K. and Mostowski, A. Set Theory, North Holland P.C., Amsterdam, 1967

Meissner, G., “Trading Financial Derivatives”, Pearson 1998

Meissner. G., “Arbitrage Opportunities in Fixed Income Markets” Derivatives Quarterly, April
1999

Merton R., “Theory of Rational Option Pricing”, Bell Journal of Economics and Management
Science, Spring 1973, 4(1), p.141-183
Vasicek, O., “An Equilibrium Characterization of the Term Structure,” Journal of Economics, 5,
1977, p.177-188
Wigner, E. P., (1932) On the quantum correction for thermodynamic equilibrium, Phys. Rev. v.
40, pp. 749-759

[Type text]