One of the key questions with setting up the Merton model, or any structural model, is where the point of default, or “knock-out,” occurs. Conceptually, the model is set up on the belief that once the value of the company's assets drops far below the amount of debt that needs to be repaid, the company will default on its debt as continued operation is unlikely to have value to the equity holders. The “barrier” level can be set in any number of ways, but the goal in setting a barrier level is to calibrate the model to the observed market.
In Merton's initial paper, he set the barrier to 100 percent of a company's short-term liabilities (due in a year or less) plus 50 percent of a company's long-term liabilities (due in a year or more). The reasoning behind this was that the management and equity of a company, as holders of an option on the assets of the firm, would tolerate the value of a firm's assets being below the level of liabilities to a certain extent. In these cases where the equity is “out of the money” or “under water,” the equity still can have substantial option value, especially in volatile or highly leveraged industries.
Given Merton's influence on the world of quantitative finance, there have been many papers published in peer-reviewed journals discussing the correct calibration of the model to observed defaults. Some of these papers segregate companies by industry and others separate them by the term structure of their debt. The initial assumption that Merton used is usually taken as a reasonable starting point.
Now that we have the more static components of our model set, we need to determine the two crucial variables, V and σv. If you will recall in the original Black-Scholes model we used for equity options, we were able to calculate the implied volatility of a stock (σ) based on the price of the option (C for a call), the price of the stock (S), and the strike price (K).
In a structural model, the value of a firm's equity is equivalent to the value of the call, as we are considering the value of the equity to be a call option on the firm's assets. We are also setting the strike price to be the face value of the debt (not the barrier, which does not have an exact equivalent in the Black-Scholes equation). However, the value of the firm's assets and the volatility of these assets are not immediately visible. This presents us with a problem in that we now have two differential equations to solve for two variables that are dependent on each other.
In our application of the Merton Model, we calculate V and σv through an iterative process off of the historical equity volatility of the stock, which we covered in our Black-Scholes example in Chapter 4. This is modeled on the method that Moody's-KMV reports that they use, though we do not have complete clarity as to the exact assumptions that they use in their proprietary models.
Once we have the model set for the company, we can proceed with the simulation. The way we set up the model is with a knock-out barrier option: Once the value of the assets drops below a certain level, the company is considered to have defaulted. This is where the simulation is done. The asset value of the firm is assumed to move with similar volatility to what has previously been calculated, and can be forecast in daily, weekly, monthly, or annual steps. See Figure 5.3
Given the complexity of structural models, we will put together a fairly basic version where we iterate to find many of the key variables, and move toward a model where the variables are calculated using market information.
Additionally, in column E enter the returns for the benchmark for this stock such as the S&P 500. We may use this to calculate an expected return through the Capital Asset Pricing Model (often abbreviated CAPM).
Above this, in E17, we will count the number of periods. Enter the formula “=COUNT(B20:B1000)”. We will reference this later. See Figure 5.4.
Below this, enter a cost of default and impair asset assumption in I6. This does not have an impact on the probability of default; it is solely for calculations around the recovery rate. At default, a company may have expenses that can run into the tens of millions of dollars. In addition, certain assets on the company's balance sheet such as goodwill and brand intangibles may be permanently lost.
=INDEX($E$6:$E$13,MATCH(B20,$B$6:$B$13,1))
With this value we can calculate the market value of the equity in each period in column G: in G20 enter “=F20*C20”. Extend these formulas down.
=INDEX($C$6:$C$13,MATCH(B20,$B$6:$B$13,1))
+INDEX($D$6:$D$13,MATCH(B20,$B$6:$B$13,1))
This adds short-term and long-term liabilities together to get the face value of all liabilities. In I20 calculate the value of the default barrier by including the weightings in I4 and I5 into the equation.
=$I$4*INDEX($C$6:$C$13,MATCH(B20,$B$6:$B$13,1))
+$I$5*INDEX($D$6:$D$13,MATCH(B20,$B$6:$B$13,1))
Extend these formulas down.
Sub iterateMerton() Range(“K20:K1000”).Value = Range(“J20:J1000”).Value Do While Range(“sumSquaredErrors”) > 10 ^ −4 Range(“K20:K1000”).Value = Range(“M20:M1000”).Value Loop End Sub
Note that we haven't created the range “sumSquaredErrors” yet; we will do this in a few steps. For now, copy the cells in J20:J1000 and paste values in K20:K1000 so we have values for the first step. This isn't necessary but will help you visualize how the spreadsheet will work.
Function BlackScholesD1(assetVal, strikePrice, rfRate, assetVol, timeMat) BlackScholesD1 = (Log(assetVal / strikePrice) + (rfRate + 0.5 * assetVol ^ 2) * timeMat) / (assetVol * timeMat ^ 0.5) End Function
Now to actually calculate the asset value in each period we will create the current period's asset values in column M. In M20 we will enter the Black-Scholes formula to calculate asset values. Enter the following (and be careful to handle the parentheses correctly):
=(G20+H20*EXP(−D20*1)
*NORMSDIST(BlackScholesD1(K20,I20,D20,$L$18,1))−
$L$18)/NORMSDIST(BlackScholesD1(K20,I20,D20,$L$18,1))
This is a rearrangement of equation 5.1 in order to solve for the asset value V.
E17: numDates
D20: riskFreeRate
G20: equityValue
H20: bookLiabilities
I20: defaultBarrier
K20: finalAssetValues
L4: Asset Volatility: “=L18”
L5: Asset Value: “=OFFSET(finalAssetValues,numDates-1,0)”
L6: Liability Value: “=OFFSET(bookLiabilities,numDates-(1,0)”
L7: Sum of Squared Errors: “=SUMXMY2(K20:K1000,M20:M1000)”.
Be sure to name this range “SumSquaredErrors” so the
“iterateMerton()” macro references it.
L8: Drift: we will come back to the drift calculation in a moment.
Leave this blank for now.
L9: Default Barrier: “=OFFSET(defaultBarrier,numDates-1,0)”
L10: Distance to Default: “= (LN(L5/L9)+(L8 - (L4^2)/2))/L4”.
This is equation 5.4.
L11: One Year Default Probability: “=NORMSDIST( - L10)”
L12: Current Liabilities: “=(OFFSET(bookLiabilities,numDates-1,0))”
L13: Estimated Recovery: “=(OFFSET(defaultBarrier,numDates-1,0)- I6)/L12”
As of 2010, in addition to constructing your own model using the previous steps, a number of companies provide structural models for default analysis. Moody's-KMV, which has been previously mentioned, was one of the first commercial outfits to make structural model results publicly available. KMV is an acronym made from the last names of Stephen Kealhofer, John McQuown, and Oldrich Vasicek, who started their company to provide extensions of the Merton Model in 1989 and sold it to Moody's in 2002.
Bloomberg also made a structural model available to its subscribers in 2010. This model is available by using the CRAT command for most public companies. Like Moody's-KMV, Bloomberg calculates a distance to default and then assigns a rating to different default probabilities. While Bloomberg does provide some documentation for its model, it is unclear how much back-testing has been done to check the calibration of the model. Similarly, MSCI, a spinoff of Morgan Stanley, also indicates that it provides a structural model of credit risk. This model is also bundled within other products that have been developed under the Barra brand name. Additionally, a number of small consulting services provide customized credit analysis work under the structural model family of analysis.