The mathematics of Ponzi schemes

Marc Artzrouni
Department of Mathematics, University of Pau, 64000, FRANCE

Abstract

A first order linear differential equation is used to describe the dynamics

of an investment fund that promises more than it can deliver, also known
as a Ponzi scheme. The model is based on a promised, unrealistic interest
rate; on the actual, realized nominal interest rate; on the rate at which new
deposits are accumulated and on the withdrawal rate. Conditions on these
parameters are given for the fund to be solvent or to collapse. The model
is fitted to data available on Charles Ponzi’s 1920 eponymous scheme and
illustrated with a philanthropic version of the scheme.
Key words: Ponzi scheme, Charles Ponzi, differential equation, investment

1. Introduction

On December 11, 2008, Bernard L. Madoff was arrested for allegedly

running a multibillion-dollar Ponzi scheme in which investors across the world
lost $ 50 billion dollars (Henriques and Kouwe, 2008). The principle of such
a scheme is simple: you entice potential investors by promising a high rate
of return, which you cannot possibly deliver. The only way you can pay the

Email address: marc.artzrouni@univ-pau.fr (Marc Artzrouni)

URL: www.univ-pau.fr/∼artzroun (Marc Artzrouni)

Preprint submitted to Mathematical Social Sciences April 1, 2009

promised interest is by attracting new investors whose money you use to pay
interests to those already in the fund.
Various macroeconomic models of rational Ponzi games have been de-
scribed in the literature (O’Connell and Zeldes, 1998; Bhattacharya, 2003;
Blanchard and Weil, 2001; Forslid, 1998). Micro Ponzi schemes have prolif-
erated particularly on the Internet to the point that regulators fear ”Ponzi-
monium” (Blas, 2009). Here we wish to investigate the mathematics of Ponzi
schemes by going beyond the simplistic pyramid-type explanations that rely
on a more or less rapid doubling of the number of new investors.
Madoff’s fund is only the most recent and perhaps biggest Ponzi scheme in
history. Madoff offered a supposedly safe 10 % return, which was considered
unrealistic, particularly in the financial climate prevailing at the end of 2008.
Still, if he could invest the money at a two or three percent interest rate and
if new deposits came in fast enough he could use the fresh money to pay
those who withdrew their earnings. But how fast must new deposits come
in? How long can the scheme last? What parameters drive the dynamics of
the fund?
In this paper we provide some answers to these questions with a simple
model that captures the main features of a Ponzi scheme. The model is
described in section 2 with detailed results on the behavior of the fund as
a function of seven parameters. In section 3 we describe Charles Ponzi’s
eponymous 1920 scheme and crudely fit the model to the data available (see
Zuckoff (2005) and Dunn (2004) for biographies of this colorful character).
We also discuss an improbable “philanthropic Ponzi scheme” that may have
something to do with Social Security. Highlights and the main results are

2
brought together in the concluding section 4.

2. The model

2.1. Assumptions
We assume that the fund starts at time t = 0 with an initial deposit K ≥ 0
followed by a continuous cash inflow s(t). We next assume a promised rate
of return rp and a nominal interest rate rn at which the money is actually
invested. If rn ≥ rp then the fund is legal and has a profit rate rn − rp . If
rn < rp the fund is promising more than it can deliver. The promised rate
rp may be called the “Ponzi rate” and is equal to 0.10 in the introductory
example. The nominal interest rate rn is 0.02 or 0.03 in that example.
We need to model the fact that investors withdraw at least some of their
money along the way. The simplest way of doing this is to assume a constant
withdrawal rate rw applied at every time t to the promised accumulated cap-
ital. The withdrawal at time t by those who invested the initial amount K is
rw Ket(rp −rw ) . If rw is less than the promised rate rp then these withdrawals
increase exponentially; rw can also be larger than rp in which case with-
drawals decrease exponentially as these investors are eating into the capital
K.
In order to calculate the withdrawals at time t from those who added to
the fund between times 0 and t we note that those who invested s(u) at time
u will want to withdraw at time t > u a quantity rw s(u)e(rp −rw )(t−u) . Inte-
grating these withdrawals from 0 to t and adding the previously calculated
withdrawals from the initial deposit K yields the total withdrawals at time t
µ Z t ¶
def. t(rp −rw ) (rp −rw )(t−u)
W (t) = rw Ke + s(u)e du . (1)
0

3
We note that the nominal interest rate rn does not appear in W (t): with-
drawals are based only on the promised rate of return rp .

2.2. The differential equation

If S(t) is the amount in the fund at time t then S(t + dt) is obtained by
adding to S(t) the nominal interest rn S(t)dt, the inflow of fresh money s(t)dt
and subtracting the withdrawals W (t)dt:

S(t + dt) = S(t) + dt[rn Sa (t) + s(t) − W (t)]. (2)

For dt → 0 the amount S(t) is the solution to the first order linear differential
equation
S(t)
= rn Sa (t) + s(t) − W (t). (3)
dt
We let C = S(0) be the initial condition which may or may not be equal to
K, the initial deposit made by customers. The fund managers can make an
initial “in-house” deposit K0 ≥ 0, which will also be invested at the nominal
rate rn . In this case the initial value C = K0 + K is larger than K. An initial
condition C < K formally corresponds to the case where for some reason
(theft or other) a fraction of the initial deposits K is not available. We will
see later that the solution to the differential equation with an initial condition
C = S(0) other than K will be used when there is at some subsequent time
t∗ a sudden change in parameter values. (For example the cash inflow or
withdrawal rate changes at t∗ ).
A simple assumption one can make on the cash inflow s(t) is that of
exponential growth:
s(t) = s0 eri t (4)

4
where s0 is the initial density of the deposits and ri will be called the invest-
ment rate. The withdrawals function W (t) of Eq. (1) is now
µ ¶
t(rp −rw ) et(rw +ri −rp ) − 1
W (t) = rw e K + s0 (5)
rw + ri − rp

where the fraction should be taken equal to t when rw + ri − rp = 0.

The solution S(t) to the differential equation (3) has a closed-form ex-
pression which will be formulated using the function

def.
g(t, a, b, c, d, α) = aebt + cedt + α. (6)

With this notation S(t) is

S(t) = g(t, a, b, c, d, α)ern t = ae(b+rn )t + ce(d+rn )t + αern t . (7)

where
def. rw [s0 − (ri − rp + rw )K]
a = , (8)
(rp − rn − rw )(ri − rp + rw )
def.
b = rp − rn − rw , (9)
def. s0 (ri − rp )
c = , (10)
(ri − rn )(ri − rp + rw )
def.
d = ri − rn , (11)
def. s0 (rn − rp ) + Krw (ri − rn )
α = C− . (12)
(ri − rn )(rn − rp + rw )
The solution S(t) of Eq. (7) is a linear combination of three exponentials,
which we were not able to tackle directly by elementary methods. Indeed,
the zeros of S(t) and of its derivative can be calculated numerically but no
closed forms were found, which precludes analytical results on the behavior
of the function.

5
 If however we know the number of positive zeros of S(t) we can shed light
on the conditions under which the fund is solvent (S(t) remains positive).
We will see that depending on parameter values S(t) of Eq. (7) has 0, 1 or
2 positive zeros. When there is no positive zero then S(t) remains positive
and the fund is solvent. One positive zero means that S(t) becomes negative
and the fund has collapsed. Two positive zeros mean that S(t) becomes
negative, reaches a negative minimum, then becomes positive again. The
fund has collapsed but could recover with a bailout equal to the absolute
value of the negative minimum. (See Bhattacharya (2003) for an economist’s
bailout model of a Ponzi scheme). To simplify we will say in this case that
the fund has collapsed then recovered.
Analytical results on the number of positive zeros will be obtained by not-
ing that the zeros of S(t) are also those of g(t, a, b, c, d, α) of Eq. (6). This
function is a linear combination of only two exponentials plus a constant. The
zeros still cannot be found in closed form by elementary methods. However
the derivative of g(t, a, b, c, d, α) is a linear combination of two exponentials
with no constant, which can be studied analytically. The following proposi-
tion provides results on the number of positive zeros of g(t, a, b, c, d, α).

Proposition 1. We consider the function g(t, a, b, c, d, α) of Eq. (6) in the

non-trivial case a, b, c, d 6= 0 and b 6= d. We also assume that g(0, a, b, c, d, α) ≥
0. We consider the following set of two conditions:

def. cd def. 1 + cd/ab

U = < 0, V = < 0. (13)
ab b−d

6
The function g(t, a, b, c, d, α) has an extremum

µ ¶ b µ ¶ d
def. −cd b − d −cd b − d
m = a +c +α (14)
ab ab

at the positive value µ ¶

−cd
ln
def. ab
tc = (15)
b−d
if and only if Condition (13) is satisfied.
We have the following results, broken down into fours cases A1 , A2 , A3 , A4 ,
on the number of positive zeros of g(t, a, b, c, d, α):
A1 : Condition (13) is satisfied and ab+cd < 0. If m > 0 then g(t, a, b, c, d, α)
has no positive zero (and therefore remains positive). If m < 0 and b, d, α < 0
then the function g(t, a, b, c, d, α) has exactly one positive zero at a value
smaller than tc . In all other cases with m < 0 the function has one positive
zero on each side of tc .
A2 : Condition (13) is satisfied and ab + cd > 0. If b, d < 0, α > 0 then
the function has no positive zero. In all other cases there is one positive zero.
A3 : Condition (13) is not satisfied and ab + cd < 0. If b, d < 0, α > 0,
the function has no positive zero. In all other cases there is one positive zero.
A4 : Condition (13) is not satisfied and ab + cd > 0. There is no positive
zero.

Proof. The proof is elementary and hinges on the following facts:

1. The derivative g 0 (t, a, b, c, d, α) equals 0 at tc and m is the value of

g(t, a, b, c, d, α) at tc ; tc is positive if and only if Condition (13) is sat-
isfied. The derivative g 0 (0, a, b, c, d, α) at 0 is equal to ab + cd.

7
 2. If b and d are negative the function g(t, a, b, c, d, α) tends to α for t →
∞. If b or d is positive the function tends to ±∞ (depending on the
signs of a, c).
3. If Condition (13) is not satisfied then either g(t, a, b, c, d, α) has an
extremum for a negative value of t or no extremum at all. In both
cases the function g(t, a, b, c, d, α) for t > 0 is monotone increasing if
ab + cd > 0 and monotone decreasing otherwise.

The limiting cases ab + cd = 0, m = 0, etc. pose no difficulty and are left

as exercises.

2.3. Main result

In order to apply Proposition 1 to the parameters a, b, c, d, α of Eqs. (8)-

(12) we first define

def. def. Krw

ρ = ri − rp , σK = − 1. (16)
s0

We will need the function

def. s0 (rn − rp ) + Krw (ri − rn )

C1 (K) = (17)
(ri − rn )(rn − rp + rw )

which is the critical value of C above which α of Eq. (12) is positive.

We define the function
K
def. s0 (rw + ri − rp ) − 1
Z(K) = , (18)
(ri − rp )/rw

and note that Z(K) = 1 if and only if K = s0 /rw (i.e. σK = 0).

8
 The extremum m of Eq. (14) and the corresponding tc of Eq. (15) are
rn − ri
(rp − ri )Z(K) rp − ri − rw
m = s0 + C − C1 (K), (19)
(ri − rn )(rn − rp + rw )

ln (Z(K))
tc = . (20)
rw + ri − rp
We also define the function C2 (K) as
rn − ri
def. (rp − ri )Z(K) rp − ri − rw
C2 (K) = C1 (K) + s0 if K ≥ s0 /rw , (21)
(ri − rn )(rn − rp + rw )
def.
C2 (K) = 0 if K < s0 /rw . (22)

The quantity C2 (K) of Eq. (21) is the critical value of C above which
the extremum m of Eq. (19) is positive.
With these notations we have the following result on the number of pos-
itive zeros of S(t) of Eq. (7).

Theorem 1. We consider the solution S(t) of Eq. (7) defined by the non-
negative parameters K, C, s0 , ri , rw , rp and rn . The number of positive zeros
of S(t) is given as a function of the sign of ρ (Figure 1):
Case B1 : ρ > 0, (ri > rp ).
Sub-case B1,1 : σK < 0 (K < s0 /rw ). S(t) has no positive zero.
Sub-case B1,2 : σK > 0 (K > s0 /rw ). We first consider the case rn > ri .
If C > C2 (K) (which includes the case C = K) then S(t) has no positive
zero for t > 0 and therefore remains positive for all t > 0. For C1 (K) <
C < C2 (K) the function S(t) has one positive zero on each side of tc . For
C < C1 (K) the function S(t) has one positive zero. When rp < rn < ri the

9
function S(t) has one positive zero for C < C2 (K) (which includes the case
C = K) and none if C > C2 (K). When rn < rp the function S(t) has one
positive zero for C < C2 (K) (which includes the case C = K if K is larger
than the fixed point K ∗ = C2 (K ∗ ) of C2 (K)) and none if C > C2 (K) (which
includes the case C = K if K is smaller than the fixed point K ∗ ).
Case B2 : ρ < 0, (ri < rp ).
Sub-case B2,1 : rw < rp −rn or rn < ri . The function S(t) has one positive
zero.
Sub-case B2,2 : rw > rp − rn and rn > ri . For C > C1 (K) (which includes
the case C = K if rn > rp ) the function S(t) has no positive zero. For
C < C1 (K) (which includes the case C = K if rn < rp ) then S(t) has one
positive zero.

Proof. The application of Proposition 1 hinges on the following observations:

1. The parameter α of Eq. (12) is positive if and only if C > C1 (K).

2. The extremum m of Eq. (14) is positive if and only if C > C2 (K).
3. The difference C2 (K) − C1 (K) has the same sign as (rp − ri )/[(ri −
rn )(rn − rp + rw )].
4. For ρ > 0 the function C2 (K) is an non decreasing function of K > 0
that has no positive fixed point if rn > rp and one positive fixed point
K ∗ = C2 (K ∗ ) if rn < rp .
5. The parameter σK and the derivative

g 0 (0, a, b, c, d, α) = ab + cd = s0 − Krw = −s0 σK (23)

of g(t, a, b, c, d, α) at 0 have opposite signs.

10
With a, b, c, d, α of Eqs. (8)-(12) the quantities U and V of (13) are

ρ σK
U= , V = , (24)
−rw σK − (σK + 1)ρ −rw σK − (σK + 1)ρ

and are both negative if and only if ρ and σK have the same sign (because
σK + 1 > 0).
In the case B1 (ρ > 0) we consider two sub-cases.
Sub-case B1,1 : σK < 0. Condition (13) is not satisfied and the derivative
of g(t, a, b, c, d, α) at 0 is positive. The result follows from A4 of Proposition
1.
Sub-case B1,2 : σK > 0. Condition (13) is satisfied and the derivative of
g(t, a, b, c, d, α) at 0 is negative. The results follow from A1 of Proposition 1.
In both sub-cases of B2 (ρ < 0) the proof relies on the sign of σK . When
σK < 0 Condition (13) is satisfied and the derivative of g(t, a, b, c, d, α) at 0
is positive. The sub-cases B2,1 and B2,2 correspond to b = rp − rn − rw or
d = ri − rn positive and to b and d negative, respectively. The results follow
from A2 of Proposition 1. When σK > 0 Condition (13) is not satisfied and
the derivative of g(t, a, b, c, d, α) at 0 is negative. The results follow from A3
of Proposition 1.

2.4. Interpretation of results

Theorem 1 breaks down the results depending on whether the rate ri of

new investments is larger or smaller than the promised rate of return rp .
We first consider the case when ri is larger than rp (Case B1 ) and the
nominal rate of return rn is also larger than rp (legal fund). In the sub-
case B1,1 (K < s0 /rw ) the fund is solvent (Z = 0) regardless of the initial

11
 1
Case B1 : ρ > 0 (ri > rp )

B1,1 : σK < 0 (K < s0 /rw )

B1,2 : σK > 0 (K > s0 /rw )

Ponzi Legal Legal

rw (withdrawal rate)
C C C
C 2 (K)
C 2 (K)
C 1 (K)
Z =0 K ∗ Z =0 Z =0
C 2 (K) Z =2
Z =1 Z =1 Z =1
s0 /rw K s0 /rw K s0 /rw K

rp ri
0
0 rn (nominal rate) 1

1
Case B2 : ρ < 0 (ri < rp )

B2,1 B2,2

Ponzi Ponzi Legal

(Philanthropic
if C > C 1 (K))
rw (withdrawal rate)

Z =1 C C
C 1 (K)
C 1 (K)
Z =0 Z =0
rp
Z =1 Z =1

K K
ri

ri rp
0
0 rn (nominal rate) 1

Figure 1: Number Z of zeros of the function S(t) in the cases B1 (ρ > 0) and B2 (ρ < 0);
Z is given in the phase space (rn , rw ). In the sub-case B2,1 the number Z of zeros is 1
regardless of the values of C and K. In the two sub-cases of B2,2 and in the three regions
of B1 , the number Z is given in the phase space (K, C).
12
condition C. In the sub-case B1,2 (K > s0 /rw ) the two rightmost graphs
in Figure 1 (Case B1 ) show that the fund remains solvent when C remains
above C2 (K) (which includes the case K = C because the first diagonal
(dash-dot line) is above C2 (K)). For rn between rp and ri the fund collapses
(Z = 1) as soon as C drops below C2 (K) (second graph). For rn larger than
ri the fund collapses but recovers (Z = 2) if C does not fall too much below
C2 (K) (C1 (K) < C < C2 (K), third graph). If C is too small (C < C1 (K))
then the fund collapses (Z = 1).
The first graph of Figure 1, Case B1 , shows what happens in a Ponzi
scheme (rn < rp ) even with a rate of new investments ri larger than rp .
The fund will be solvent for C = K only if K is not too large (K less than
the fixed point K ∗ ). If C = K and is larger than the fixed point K ∗ , then
the combined withdrawals by the initial and subsequent investors eventually
cause the fund to collapse.
In Case B2 (the rate of new investments ri is smaller than the promised
rate of return rp ) we first consider the sub-case of B2,2 with rn > rp (legal
fund, second graph). The fund remains solvent for C > C1 (K), which in-
cluded the case C = K. In the Ponzi sub-case of B2,2 with ri < rn < rp and
rw > rp − rn the fund does not grow too fast and is solvent if C is larger than
C1 (K), which is itself larger than K (first graph). This means that despite
an ri and an rn smaller than rp , the Ponzi scheme is solvent if the fund man-
ager can add to K an“in-house” investment K0 at least equal to C1 (K) − K.
We will see in the numerical illustrations that C1 (K) − K can be quite large
and the scheme unprofitable for the fund manager, hence the “philanthropic
Ponzi scheme” characterization. We emphasize for future reference that this

13
scenario hinges on an investment rate ri that remains smaller than the nom-
inal rate rn . If the manager does not invest enough (C < C1 (K)) the fund
collapses.
The Ponzi sub-case B2,1 consists of the values rn < ri and of the values
(rn , rw ) for which rw < rp − rn and rn is between ri and rp (triangular region
below the first graph). In this sub-case B2,1 the fund grows too fast and
collapses (Z = 1).
This analysis shows that the role of rw is ambiguous when rn is between
ri and rp . Although a small rw (rw < rp − rn , B2,1 ) may seem desirable, the
fund will grow more in the long run and eventually collapses. A large rw
(rw > rp − rn , B2,2 ) may seem dangerous but depletes the fund and means
smaller withdrawals in the long run. The fund is ultimately solvent if C
is large enough to absorb the large early withdrawals (“philanthropic Ponzi
scheme”).

2.5. Actual and promised amounts in the fund

In order to describe the dynamics of a fund that includes a sudden pa-

rameter change at some time t∗ , we need to make explicit the role of the
parameters by denoting S(t, K, C, s0 , ri , rw , rp , rn ) the solution of the differ-
ential equation given in (7).
We introduce the actual and promised amounts Sa (t) and Sp (t). The
actual amount Sa (t) in the fund (based on the nominal rate of return rn and
the initial condition C), is the one given in Eq. (7) and rewritten explicitly
as

rw [s0 − (ri − rp + rw )K] (rp −rw )t

Sa (t) = S(t, K, C, s0 , ri , rw , rp , rn ) = e +
(rp − rn − rw )(ri − rp + rw )

14
 µ ¶
s0 (ri − rp ) ri t s0 (rn − rp ) + Krw (ri − rn ) rn t
e + C− e (25)
(ri − rn )(ri − rp + rw ) (ri − rn )(rn − rp + rw )
The quantity Sp (t) is the amount that is promised to and belongs to
investors; Sp (t) is obtained by setting in Eq. (25) the parameter rn equal to
rp and the initial condition C equal to K. Under these conditions the third
term in Eq. (25) becomes zero and

Sp (t) = S(t, K, K, s0 , ri , rw , rp , rp ) =

s0 ¡ (rp −rw )t ¢
e − eri t + Ke(rp −rw )t (26)
rp − ri − rw
Contrary to the actual amount Sa (t) in the fund the promised amount Sp (t)
is positive regardless of the parameter values.

2.6. Change in parameter values

The assumption of an exponentially increasing density of new investments

is not realistic in the long run and we may wish to examine what happens
if the investment rate ri suddenly drops to 0. This means that the flow of
new investments becomes a constant. More generally, it would be useful to
be able to describe the future dynamics of the fund if at a point in time t∗
the parameters (s0 , ri , rw , rp , rn ) experience a sudden (discontinuous) change
of value and become (s00 , ri0 , rw0 , rp0 , rn0 ).
We call C 0 and K 0 the actual and the promised amounts in the fund at
time t∗ :
C 0 = Sa (t∗ ) = S(t∗ , K, C, s0 , ri , rw , rp , rn ) (27)

and
K 0 = Sp (t∗ ) = S(t∗ , K, K, s0 , ri , rw , rp , rp ). (28)

15
These quantities will be the new initial condition and initial investment start-
ing at time t∗ . The actual and promised amounts at any time t are now

 S(t, K, C, s , r , r , r , r ) if t ≤ t∗ ;
0 i w p n
Sa (t) = (29)
 S(t − t∗ , K 0 , C 0 , s0 , r0 , r0 , r0 , r0 ) if t > t∗ .
0 i w p n

and 
 S(t, K, K, s , r , r , r , r ) if t ≤ t∗ ;
0 i w p p
Sp (t) = (30)
 S(t − t∗ , K 0 , K 0 , s0 , r0 , r0 , r0 , r0 ) if t > t∗ .
0 i w p p

Several discontinuous parameter changes at different times can be dealt with

in this fashion.

3. Applications

3.1. Charles Ponzi’s original scheme

Ponzi schemes are named after the eponymous Italian adventurer and
fraudster Charles Ponzi (1882-1949) who emigrated to the United States in
1903. We will use the data available on Ponzi’s scheme to fit crudely the
model and study the implications of a density of new deposits becoming
constant (ri = 0) at some point t∗ . The data used comes from DeWitt
(2009).
Charles Ponzi was a petty crook who spent time in and out of prison.
While in Boston in 1919 he learnt that International Reply Coupons could
be bought in one country and theoretically resold in another at a profit.
Trading in these coupons was not realistic, but that did not prevent Ponzi
from issuing bonds which offered a 100% profit if held for 90 days (0.25 years).
We translate this 90-day doubling time into an annualized rate of return rp
that must satisfy e0.25rp = 2, i.e. rp = 2.773.

16
 20
Promised amount Sp (t)
18

16

14
Dollars (millions)

12

10 Actual amount Sa (t)

2
0.58
0
0 0.5 1 1.5
Years

Figure 2: Actual and promised amounts Sa (t) and Sp (t) in Ponzi’s fund, crudely fitted
to available data for the duration of the fund from time t = 0 (December 26, 1919) to
t∗ = 0.58 (July 26, 1920). The hypothetical trajectories beyond 0.58 illustrate the effect of
the flow of new investments becoming constant and equal to the $ 200,000 a day reached
at time 0.58. The fund would have collapsed about nine months later (Sa (1.30) = 0).

Ponzi was penniless when he launched his fund and there is no reported
initial investment K, so we can safely set C = K = 0. The parameters s0 and
ri for the density of new investments s0 exp(ri t) can be estimated crudely on
the basis of information on deposits made between the first day (December
26, 1919, t = 0) and last day (July 26, 1920, t∗ = 0.58) of the scheme’s
history. On the last day Ponzi collected $200,000 for a total of $10 million
deposited over the seven-month period. Under the assumption of exponential
growth and with the year and a million dollars as the time and monetary

17
units, the parameters s0 and ri must satisfy
µ ¶
0.58ri e0.58ri − 1
s0 e = 0.2 × 365, s0 = 10 (31)
ri

which yields
s0 = 1.130, ri = 7.187. (32)

This value of s0 translates into an initial flow s0 /365 of $ 3,095 a day. An

exponential growth from $ 3,095 to $ 200,000 a day seems plausible in view
of Ponzi’s own description of A huge line of investors, four abreast, stretched
from the City Hall Annex . . . all the way to my office! (quote in DeWitt
(2009)).
The actual profit made on the International Reply Coupons was negligi-
ble. We thus arbitrarily choose a nominal rate rn of 0.01, although rn could
just as well be set equal to 0. Indeed, the impact of a one or even five percent
nominal return is negligible in comparison with new investments pouring in
at an instantaneous rate of 7.187.
The greatest uncertainty rests with the redemption rate (our withdrawal
rate rw ) at which investors cashed in the bonds. We only know that some
investors redeemed their bonds after 90 days but that many left their money
to double once again. We can however estimate rw crudely on the basis of
information available at the time of Ponzi’s trial. He redeemed $5 million
of his bonds after the July 1920 cessation of activity with $7 million still
outstanding. We will take these figures to mean that the promised amount
accumulated by time t∗ = 0.58 is

Sp (0.58) = 12. (33)

18
 K C s0 ri rw rp rn
Ponzi’s scheme (t ≤ 0.58) 0 0 1.130 7.187 1.47 2.773 0.01
Ponzi’s scheme (t > 0.58) 12 7.76 73 0 1.47 2.773 0.01
Philanthropic Ponzi scheme 0 280 1 0 0.12 0.15 0.04

Table 1: Model parameter values for Charles Ponzi’s 1920 scheme (Section 3.1) and for a
hypothetical philanthropic Ponzi scheme (Section 3.2).

With all other parameter values known, we solved this equation numerically
for rw and obtained rw = 1.47 (Table 1). The corresponding actual amount
in the fund at t∗ = 0.58 is then Sa (0.58) = $7.763 million. After the late
$5 million redemption Ponzi was left with $2.763 million, which is consistent
with his reported assets of $2 million at the time of his trial if we assume he
had spent $0.763 million.
Figure 1 shows that with rn < rp < ri and K = C = 0 the scheme
corresponds to the first graph of Case B1 and to the sub-case B1,1 . The
function Sa (t) has no zero and would continue to grow at the asymptotically
exponential rate ri (second term of Eq. (25)) as long as investors pour money
in at the same rate ri .
Figure 2 depicts the actual and promised amounts up to t∗ = 0.58 on
the basis of the parameter values estimated above. As a projection exercise
we assume that after t∗ = 0.58 the flow of new deposits stopped increasing
and remained equal to the $200,000 per day reached at t∗ = 0.58 (July 26,
1920). This means that the annualized density is now s00 = 200, 000 × 365 =
$73 million a year and that ri became ri0 = 0. The other parameter values
were kept unchanged after t∗ = 0.58 (rn0 = 0.01, rp0 = 2.773, rw0 = 1.47, Table

19
1). With ri0 = 0 < rn0 = 0.01 < rp0 = 2.773 we are now in the sub-case
B2,1 of B2 because rw0 = 1.47 < rp0 − rn0 = 2.763. As expected the fund
would have collapsed, with Sa (t) reaching 0 roughly nine months after t∗
(Sa (1.30) = 0). The promised amount Sp (t) continues to grow exponentially
at the rate rp0 − rw0 = 1.303.

3.2. A philanthropic Ponzi scheme

We noted that if rw > rp − rn and ri < rn < rp , then for C > C1 (K)
the Ponzi scheme is solvent. We will now see why we labeled this scenario
“philanthropic”. With C − C1 (K) positive, the asymptotic growth rate of
the actual amount Sa (t) is rn because the last term (C − C1 (K))ern t in Eq.
(25) dominates all other exponential terms. The promised amount Sp (t) of
Eq. (26) grows at the asymptotic rate rp − rw which is smaller than the
nominal rn . The fund’s profit is Sa (t) − Sp (t) and we have for large t

Sa (t) ∼ Sa (t) − Sp (t) ∼ (C − C1 (K))ern t (34)

where x(t) ∼ y(t) means that lim x(t)/y(t) = 1. If the initial deposit is
t→∞
K = 0 then the initial condition C is equal to the“in-house” investment K0
and (34) becomes
µ ¶
rn t s0 (rp − rn )
Sa (t) ∼ Sa (t)−Sp (t) ∼ (C −C1 (0))e = C− ern t
(rn − ri )(rn − rp + rw )
(35)
where C must be larger than C1 (0). The actual amount Sa (t) and the profit
Sa (t) − Sp (t) grow asymptotically at the exponential rate rn but are both
smaller than the profit Cern t that the manager could have made by simply
investing the quantity C at the nominal rate rn . This shows that solvency

20
 0.06

0.055
Logarithms of amounts divided by years

0.05
ln(C)/t + rn

0.045
ln(Sa (t))/t

0.04
rn = 0.04

0.035 ln(Sp (t))/t

0.03 r − r = 0.03
p w

0.025

0.02
0 500 1000 1500 2000
Years

Figure 3: Natural logarithms divided by time t of actual and promised amounts in philan-
thropic Ponzi fund, compared with the larger natural logarithm divided by t of the accu-
mulated amount had the $ 280 million been invested at the nominal rate rn (ln(C)/t + rn ).

hinges on a philanthropic fund manager who is willing to invest a significant

initial amount and give away a share of her profits.
We will see how large the initial investment C must be with the example
of a manager offering a 15% return to individuals who contribute one million
dollars a year and continuously withdraw 12% of their accumulated capital
(rp = 0.15, s0 = 1, ri = 0, rw = 0.12, Table 1). We assume the manager
can realistically earn 4 % (rn = 0.04). With no initial deposits by investors
(K = 0) the fund manager’s minimum investment C to keep the fund solvent
is C1 (0) = 275. With C = 280 we plotted in Figure 3 the functions ln(Sa (t))/t
and ln(Sp (t))/t to show their convergence to rn = 0.04 and rp − rw = 0.03

21
respectively. We also plotted ln(Cern t )/t = ln(C)/t + rn , which is the natural
logarithm divided by time t of the accumulated amount had the $280 million
been invested at the nominal rate rn . This quantity also converges to rn =
0.04 while staying larger than ln(Sa (t))/t and ln(Sp (t))/t.
An initial investment of $280 million dollars to sustain (at a loss) a fund
that grows by only $1 million a year may not be a fund manager’s idea of a
profitable venture. This example was provided only to illustrate the fact that
with a sufficiently large initial “in-house” investment a Ponzi scheme can be
solvent with no or a very small growth in deposits and a nominal interest
rate smaller than the promised one.
The example may also contribute to the discussion about the U.S. Social
Security system being comparable to a Ponzi scheme. (See Mandel (2008) for
a balanced account). The Social Security Administration vehemently denies
any connection and suggests that the short-lived nature of Charles Ponzi’s
scheme is the telltale sign of such a fraudulent operation (DeWitt, 2009).
The same Social Security source indicates that the German and U.S. Social
Security systems have been in operation since 1889 and 1935, respectively.
DeWitt (2009) implies that this longevity disqualifies these systems as Ponzi
schemes. However our model makes clear that a Ponzi scheme can last a
long time before collapsing or even indefinitely with or without a growth
in deposits. In particular, our philanthropic Ponzi scheme which relies on
a large initial investment may approximate a perfectly legal and legitimate
non-profit government investment into a social security scheme meant to
provide fixed incomes to a growing population of retirees. We recall however
that in this case the solvency of the fund depends on an investment rate ri

22
that remains smaller than the nominal rate of return rn . Although the role
of population growth is beyond the scope of this paper, this shows that the
retiree population and its investments must not grow too rapidly in order for
the fund to be solvent.

4. Conclusion

After the collapse of Charles Ponzi’s scheme in July 1920 it took seven
years of litigation for the investors to get 37 cents on the dollar of their
principal. After several trials Ponzi spent a total of 10 years in prison before
returning to Italy where he briefly worked for Benito Mussolini. Having once
again mismanaged things he fled to Brazil where he died a pauper in a Rio
de Janeiro charity hospital in 1949.
Robbing Peter to pay Paul is an ancient practice that started long before
Ponzi and is experiencing a comeback on the Internet. This paper attempts
to add to our understanding of these schemes with the simplest possible
continuous-time mathematical model of a fund that offers more than it can
deliver. To be sure, constant investment, withdrawal and nominal interest
rates capture only crudely the variety of human behaviors and the complex-
ities of financial instruments available today. On the other hand, some of
our assumptions are quite realistic. For example there is good reason to be-
lieve that setting rn to 0 or 0.01 in the Ponzi or Madoff case is probably an
accurate description of what they did with the money.
As we expected the fund is always solvent with C = K in the case of a
legal fund characterized by rn > rp . In a Ponzi scheme (rn < rp ) the fund
can remain solvent depending on the values of the investment rate ri and the

23
withdrawal rate rw . The model sheds light on the ambiguous role played by
these two parameters. If ri is too large or rw too small the fund grows fast
and can be in jeopardy as withdrawals increase. If ri is too small or rw too
large the fund may not be able to keep up with withdrawals.
Our model yields a variety of increasing trajectories that may look alike
initially, but are fundamentally different in their long-run behavior. Some
will continue to increase as long as new investments come in - others will
increase possibly for a long time before they collapse. This happens when
parameter values in the phase spaces of solutions are close to border regions
between different qualitative behaviours (for example between no zero and
one zero for the function S(t)). In some cases S(t) initially decreases, reaches
a positive or negative minimum, and then recovers.
Finally, our results can provide concrete answers to financial regulators
and others confronted with funds that make unrealistic claims. Suppose
one investigates a fund that promises a 10% return per month, provides its
own initial “in house” investment of $5 millions dollars, claims a doubling
of its deposits every two months and has investors who withdraw half their
earnings every year. It is now possible to predict what will happen to the
fund, at least under the simple assumptions made in the model described in
this paper.

References

Bhattacharya, U. 2003. The Optimal Design of Ponzi Schemes in Finite

Economies. Journal of Financial Intermediation, 12, 2-24.

Blanchard, O. and Weil, P. 2001. Dynamic efficiency, the riskless rate, and

24
 debt Ponzi games under uncertainty, Advances in Macroeconomics, Vol 1,
Iss. 2, Article 3.

Blas, J. 2009. Watchdog fears market ”Ponzimonium”, Financial Times, 20

March 2009, Companies.

DeWitt, L. 2009. Ponzi Schemes vs. Social Security, Research Note No. 25.
Social Security Administration. http://www.ssa.gov/history/ponzi.htm.
Retrieved on March 25, 2009.

Dunn, D. 2004. Ponzi: The Incredible True Story of the King of Financial
Cons, Broadway Books, New York.

Forslid, R. 1998. External debt and Ponzi-games in a small open economy

with endogenous growth. Journal of Macroeconomics, 20,2, 341-349.

Henriques, D. and Kouwe, Z. 2008. Prominent Trader Accused of Defrauding

Clients, New York Times, 11 December, Business.

Mandel, M. 2008. Is Social Security a Ponzi Scheme?, Business Week, De-

cember 28. http://www.businessweek.com. Retrieved on March 25, 2009.

0’Connell, S.A. and Zeldes, S.P. 1988. Rational Ponzi games. International
Economic Review, vol 29(3), 431-450.

Zuckoff, M. 2005. Ponzi’s Scheme: The True Story of a Financial Legend.

Random House, New York.

25