You are on page 1of 6

The Merton Probability of Default (PD) Model

A Structural Approach to Default Prediction Implemented into Python

The Merton structural approach of evaluating default risk is a cause and effect model
because it first identifies conditions under which borrowers may default, and then it
estimates the probability that these conditions may actually occur.

Asset Value, Value of Equity, and Value of Liabilities are linked by the following
relationship:

Asset Value = Value of Equity + Value of Liabilities (1)

Mertons model has the advantage that the Probability of Default (PD) can be continually
updated with the evolution of the companys Asset Value. Its main shortcoming is that
Asset Value is unobservable and must be extrapolated from the companys share prices.

Consider default in the case of limited liability companies: Default is expected to occur if
Asset Value cannot cover the Value of Liabilities. Because equity holders receive the
residual value of the firm, the Value of Equity E is negative if the Asset Value A is
smaller than the Value of Liabilities L:

A < L then E < 0 (2)

If equity holders have something of negative value they exercise the walk-away option
because of limited liability and leave the company to the creditors. Because creditors
claims are not fully covered, the company is in default.

The walk-away option is evaluated with standard approaches from option pricing theory,
and for this reason structural models are called option-theoretic or contingent-claim
models.

Mertons set-up:
Companys Value of Liabilities L consist of one zero-coupon bond with notional value
L maturing in T;
There are no payments until T;
Equity holders will wait till T to default (or not);
The company pays no dividends.
The Probability of Default PD is then the probability that at time T the Asset Value is
below the Value of Liabilities.

To calculate PD we need:
Value of Liabilities L from balance sheet;
Probability distribution of Asset Value AT at maturity (default time) T.

A common assumption is that the logarithm of the Asset Value in T follows a normal
distribution with the following parameters:
lnAT ~ N( lnAt + ( - 2/2)(T t), 2(T t)) (3)

where is asset volatility, - 2/2 is the expected per annum change in log asset
values, is the Drift Rate, and t denotes today.

The probability that the normally distributed variable lnAT falls below ln L is given by the
cumulative standard normal distribution

2
ln AT + ( )(T t) ln L
PD = ( 2 ) (4)
T t

The Distance to Default DD measures the number of standard deviations the expected
asset value is away from default:

2
ln AT + ( )(T t) ln L
DD = 2 (5)
T t

Therefore, the Probability of Default PD is

PD = (-DD) (6)

Because we cannot observe market Asset Value, we do not know todays value At .
Furthermore, for this reason we cannot derive an estimate of Asset Volatility . This is
where option pricing theory can help because it can provide a relationship between the
unobservable (At , ) and observable variables. The market Value of Equity is given by
the share price multiplied by the number of outstanding shares if the company is publicly
traded. Using option pricing theory, at maturity T the Value of Equities ET and Asset
Value AT are then described by the formula for the pay-off for a European call option:

ET = max(0, AT L) (7)

As long as the Asset Value A is below the Value of Liabilities L, the Value of Equity E is
zero because all assets are claimed by the bondholders. If Asset Value A is higher than
the notional principal of the zero-coupon bond L, bondholders receive the residual value
of the company and their pay-off increases linearly with Asset Value A. The underlying of
the call is the companys Asset Value A, the calls strike is L. The pay-off to bondholders
corresponds to a portfolio composed of a risk-free zero-coupon bond with notional value
L and a short-put on the firms assets with strike L.

Because the company pays no dividends, the Value of Equity E can be determined with
the standard Black-Scholes call option formula:

Et = At (d1 ) L er(T t ) (d2 ) (8)



where
ln(At / L) + (r + 2 / 2)(T t)
d1 = and d2 = d1 T t (9)
T t

while r is the logarithmic risk-free rate of return. At this point we have an equation that
links the Value of Equity E at time t (an observable value) to unobservable (At , ) which
we want to determine.

One approach is to look at a one-year horizon, going back in time 260 trading days.
Rearranging the Black-Scholes formula (8):

At = [Et + L er(T t )(d2 )] / (d1 ) (10)



and going back in time 260 trading days by setting t-1, t-2, , t-260 we get a system of
equations that we can use to solve for At and . A usual assumption is made for
structural models that the company has only liabilities that mature in one year. If that is
not the case, other models are being used, not this one. However, structural models
such as the present one are useful for producing one-year default probabilities.

Under the assumption that maturity is one year, we can set (T t) to one for each of the
260 trading days. We can now implement these equations into Python as follows:

For each trading day i = 0,1,, 260 from the past we collect the market Value of Equity
Et i and the book Value of Liabilities Lt i and add them together to get Asset Value At i.
In this implementation of the model, we created a separate Python computer program to
generate 260 values for E and L for each trading day of the last year because this write-
up serves only as an example of how the structural Merton model can be implemented
into Python. An excerpt of the generated Asset Values A read back in matrix form in the
Merton model program is shown in the figure below:
Once we have all 260 Asset Values A, we compute by setting it equal to the standard
deviation of the log Asset Values A:

which returns:

We use these values for A and to start an iterative process by inserting them into the
Black-Scholes (9) and compute a new set of 260 values for At i and a new :

We continue calculating new Asset Values and a new Asset Volatility until
convergence is achieved when the sum of squared differences between consecutive
asset values is below some small value. We used the one-year US treasury as the risk-
free rate of return r.

To transform the Asset Volatility (sigma at convergence) into a per annum volatility
(which we now call asset volatility) we used the root -T rule. Results of the
computations are shown below:
To further calculate the Probability of Default PD we need the Expected Change in Asset
Value E[Ri] where Ri denotes Return on an Asset i. To determine E[Ri] we use the
Capital Asset Pricing Model (CAPM):

E[Ri] Rf = i (E[RM] Rf) (11)

where Rf is the simple risk-free rate of return

Rf = exp(r) 1 (12)

and the S&P500 Index is taken as a proxy for RM, the return on the market portfolio. By
regressing the asset value returns on the S&P500 returns, we obtain an estimate of the
slope :

Assuming a standard value of 4% for the market risk premium (E[RM] Rf), we compute
the expected asset return:

The Drift Rate is

Now that we have estimates for Asset Volatility, Asset Value and the Drift Rate, we can
compute the Distance to Default DD using (5) and the Probability of Default PD using
(4):

You might also like