Valuing Employee Stock Options Part 5 doc

This chapter focuses on the applicability of Monte Carlo simulation as it pertains to valuing stock options and as a means of simulating the inputs that go into a customized binomial lat

CHAPTER 5 Applicability of Monte Carlo Simulation

INTRODUCTION TO THE ANALYSIS

Analyses in previous chapters clearly indicate that using the BSM alone is insufficient to measure the true fair-market value of ESOs Option pricing has made vast strides since 1973 when Fischer Black and Myron Scholes published their path-breaking paper providing a model for valuing Euro-pean options While Black and Scholes’ derivations are mathematically complex, other approaches broached in this book, namely those using Monte Carlo simulation and binomial lattices, provide much simpler appli-cations but at the same time enable a similar wellspring of information.1In fact, applying binomial lattices with Monte Carlo simulation has been made much easier with the use of software and spreadsheets

This chapter focuses on the applicability of Monte Carlo simulation as

it pertains to valuing stock options and as a means of simulating the inputs that go into a customized binomial lattice—that is, used in conjunction with binomial lattices This chapter begins with a brief review of the three types of option pricing methodologies and continues with a quantitative as-sessment of their analytical robustness under different conditions The sim-ulation approach to valuing options will be shown to be precise when it comes to valuing simple European options without dividends In contrast, when it comes to American or mixed options with exotic features (vesting, forfeiture, suboptimal behavior, and blackout dates), the simulation ap-proach to valuing options breaks down and cannot be used The binomial lattice is a much better candidate when these exotic elements exist How-ever, Monte Carlo simulation still proves to be a powerful and useful tool for simulating the uncertain input variables with correlations, and allows tens of thousands of scenarios to enter into a customized binomial lattice It

is shown later in this chapter that a precision-controlled simulation can

in-51

crease the confidence of the results and narrow the errors to less than a

$0.01 precision with a 99.9 percent statistical confidence level, increasing the confidence of the valuation results

In the rest of the chapter a brief review of the three mainstream ap-proaches is made and the valuation results are then compared The simula-tion approach to valuasimula-tion will be shown not to be applicable by itself; but when coupled with the customized binomial lattices, it provides a powerful analytical tool that yields robust results

The Black-Scholes Model

In order to fully understand and use the BSM, one needs to understand the assumptions under which the model was constructed These are essentially the caveats that go into using the BSM option pricing model These as-sumptions are violated quite often, but the model still holds up to scrutiny when applied appropriately to European options A European option is the type of option that can be exercised only on its expiration date and not be-fore In contrast, most executive stock options awarded are American op-tions, where the holder of the option is allowed to exercise at any time (except on blackout dates) once the award has been fully vested

The main assumption that goes into the BSM is that the underlying as-set’s price structure follows a Brownian Motion with static drift and volatility parameters, and that this motion follows a Markov-Weiner sto-chastic process In other words, it assumes that the returns on the stock prices follow a lognormal distribution The other assumptions are fairly standard, including a fair and timely efficient market with no riskless arbi-trage opportunities, no transaction costs, and no taxes Price changes are also assumed to be continuous and instantaneous Finally, the risk-free rate and volatility are assumed to be constant throughout the life of the option, and the stock pays no dividends.2 However, for fairness of comparison, a modification of the BSM is used—the GBM This modification allows the incorporation of dividends in a standard European option

Monte Carlo Path Simulation

Monte Carlo simulation can be easily adapted for use in an options valua-tion paradigm There are multiple uses of Monte Carlo simulavalua-tion includ-ing its use in risk analysis and forecastinclud-ing Here, the discussion focuses on two distinct applications of Monte Carlo simulation: solving a stock op-tion valuaop-tion problem versus obtaining a range of solved opop-tion values Although these two approaches are discussed separately, they can be used together in an analysis

Applying Monte Carlo Simulation

to Obtain a Stock Options Value

Monte Carlo simulation can be applied to solve an options valuation prob-lem, that is, to obtain a fair-market value of the stock option Recall that the mainstream approaches in solving options problems are the binomial approach, closed-form equations, and simulation In the simulation ap-proach, a series of forecast stock prices are created using the Brownian Motion stochastic process, and the option maximization calculation is ap-plied to the series’ end nodes, and discounted back to time zero, at the risk-free rate

Note that simulation can be easily used to solve European-type op-tions, but it is fairly difficult to apply simulation to solve American-type options.3 In fact, certain academic texts list Monte Carlo simulation’s major limitation as that it can be used to solve only European-type op-tions.4If the number of simulation trials are adequately increased, cou-pled with an increase in the simulation time-steps, the results stemming from Monte Carlo simulation also approach the BSM value for a Euro-pean option

Binomial Lattices

Binomial lattices, in contrast, are easy to implement and easy to explain They are also highly flexible, but require significant computing power and lattice steps to obtain good approximations, as will be seen in the next chapter The results from closed-form solutions can be used in conjunction with the binomial lattice approach when presenting a complete stock op-tions valuation solution Binomial lattices are particularly useful in captur-ing the effect of early exercise as in an American option with dividends, vesting and blackout periods, suboptimal early exercise behaviors, forfei-tures, performance-based vesting, changing volatilities and business envi-ronments, changing dividend yield, changing risk-free rates, and so forth—the same real-life conditions that cannot be accounted for in the BSM, GBM, or simulation Binomial lattices can even account for exotic events such as stock price barriers (a barrier option exists when the stock option becomes either in-the-money or out-of-the-money only when it hits

a stock price barrier), vesting tranches (a specific percent of the options granted becomes vested or exercisable each year, or that senior manage-ment has different option grants than regular employees), and so forth Monte Carlo simulation can then be applied to simulate the probabilities

of forfeitures and underperformance of the firm, and use these as the in-puts into the binomial lattices.5

Analytical Comparison

The following presents a results comparison of the three methods dis-cussed The main goal of the analysis is to show that under certain restric-tive conditions, all three methodologies provide identical results, indicating that all three methods are robust and correct However, when conditions are changed to mirror real-life scenarios, binomial lattices provide a much more accurate fair-value assessment than the GBM and BSM approaches, where the latter approaches may sometimes overvalue and at other times undervalue the ESO

Figure 5.1 illustrates a comparative analysis of the three different op-tions valuation methodologies for a simple set of inputs The usual inputs

in the options valuation model are: expiration in years, stock price, volatil-ity, risk-free rate, dividend rate, and strike price Notice that with a simple set of inputs where the stock is assumed not to pay any dividends, the bi-nomial approach with 5,000 steps yields $39.43, identical to the BSM of

$39.43 The path simulation approach also yields a value of $39.43.6 No-tice in addition, that the American closed-form model results indicate iden-tical values when no dividend payments exist, and that all methods yield the same values in a European option In American options when a divi-dend exists, the values obtained from the three methodologies are vastly different, as seen in Table 5.1 (a–d)

When a dividend yield exists, that is, when the underlying firm’s stocks pay dividends, the results from a BSM or GBM are no longer robust or correct, because early execution is optimal, making the stock option, an American-type option, more valuable than is estimated using the BSM Table 5.1 (a–d) illustrates this point For instance, panel (a) of Table 5.1 shows the results from a BSM, and panel (b) is the binomial lattice for a European option, while the panel (c) shows the results from an American closed-form approximation model, and panel (d) shows the binomial ap-proach for an American option Notice that for all four panels, the first column results are identical when no dividends exist This indicates that all four methodologies are robust and consistent and provide identical values

at the limit, under the condition of no dividends and are all valid for Euro-pean-type options However, when dividends exist, the BSM breaks down and is no longer valid, especially when the option is of the American type

Applying Monte Carlo Simulation for

Statistical Confidence and Precision Control

Alternatively, Monte Carlo simulation can be applied to obtain a range of calculated stock option fair values That is, any of the inputs into the stock

FIGURE 5.1

Input Parameters Expiration in Y

European Option Results Binomial Approach

American Option Results Binomial Approach

Simulation Calculation Simulate V

55

TABLE 5.1 (a–d)

Binomial Approach

56

TABLE 5.1 (a–d)

Closed-Form Approximation

Binomial Approach

57

options valuation model can be chosen for Monte Carlo simulation if they are uncertain and stochastic Distributional assumptions are assigned to these variables, and the resulting options values using the BSM, GBM, path simulation, or binomial lattices are selected as forecast cells These mod-eled uncertainties include the probability of forfeiture and the employees’ suboptimal exercise behavior The simulation examples throughout this book use Decisioneering, Inc.’s Crystal Ball software

The results of the simulation are essentially a distribution of the stock option values Keep in mind that the simulation application here is used to vary the inputs to an options valuation model to obtain a range of results, not to model and calculate the options themselves However, simulation can

be applied both to simulate the inputs to obtain the range of options results and also to solve the options model through path-dependent simulation Monte Carlo simulation, named after the famous gambling capital of Monaco, is a very potent methodology Monte Carlo simulation creates ar-tificial futures by generating thousands and even millions of sample paths

of outcomes and looks at their prevalent characteristics, and its simplest form is a random number generator that is useful for forecasting, estima-tion, and risk analysis A simulation calculates numerous scenarios of a

model by repeatedly picking values from a user-predefined probability

dis-tribution for the uncertain variables and using those values for the model.

As all those scenarios produce associated results in a model, each scenario

can have a forecast Forecasts are events (usually with formulas or

func-tions) that you define as important outputs of the model

Simplistically, think of the Monte Carlo simulation approach as pick-ing golf balls out of a large basket repeatedly with replacement The size

and shape of the basket depend on the distributional assumptions (e.g., a

normal distribution with a mean of 100 and a standard deviation of 10, versus a uniform distribution or a triangular distribution) where some baskets are deeper or more symmetrical than others, allowing certain balls to be pulled out more frequently than others These balls are col-ored differently to represent their respective frequency or probabilities of occurrence The number of balls pulled repeatedly depends on the

num-ber of trials simulated For a large model with multiple related

assump-tions, imagine the large model as a very large basket, where many baby baskets reside Each baby basket has its own set of different-colored golf balls that are bouncing around Sometimes these baby baskets are linked

to each other (if there is a correlation between the variables) and the golf

balls are bouncing in tandem while others are bouncing independently of one another The balls that are picked each time from these interactions

within the model are tabulated and recorded, providing a forecast result

of the simulation

These concepts can be applied to ESO valuation For instance, the sim-ulated input assumptions are those inputs that are highly uncertain and can vary in the future, such as stock price at grant date, volatility, forfeiture rates, and suboptimal early exercise behavior multiples Clearly, variables that are objectively obtained, such as risk-free rates (U.S Treasury yields for the next 1 month to 20 years are published), dividend yield (determined from corporate strategy), vesting period, strike price, and blackout periods (determined contractually in the option grant) should not be simulated In addition, the simulated input assumptions can be correlated For instance, forfeiture rates can be negatively correlated to stock price—if the firm is doing well, its stock price usually increases, making the option more valu-able, thus making the employees less likely to leave and the firm less likely

to lay off its employees Finally, the output forecasts are the option valua-tion results

The analysis results will be distributions of thousands of options valu-ation results, where all the uncertain inputs are allowed to vary according

to their distributional assumptions and correlations, and the customized binomial lattice model will take care of their interactions The resulting av-erage (if the distribution is not skewed) or median (if the distribution is highly skewed) options value is used Hence, instead of using single-point estimates of the inputs to provide a single-point estimate of options valua-tion, all possible contingencies and scenarios in the input variables will be accounted for in the analysis through Monte Carlo simulation

Table 5.2 shows the results obtained using the customized binomial lattices based on single-point inputs of all the variables The model takes exotic inputs such as vesting, forfeiture rates, suboptimal exercise behavior multiples, blackout periods, and changing inputs (dividends, risk-free rates, and volatilities) over time The resulting option value is $31.42 This analysis can then be extended to include simulation Table 5.2 and Figures 5.2 to 5.6 illustrate the use of simulation coupled with customized bino-mial lattices.7

For instance, Figure 5.2 illustrates how the input assumptions are ob-tained through a distributional-fitting routine Using historical data, com-parable data, forecast projections, or management assumptions, the correct distributions are obtained through a rigorous statistical hypothesis test Figure 5.3 shows all the input assumptions used in the model Notice that only volatility, forfeiture rate, and suboptimal exercise behavior mul-tiple are simulated The rest of the input variables are either contractually fixed or objectively obtained (e.g., risk-free rates from the U.S Treasury) and their fluctuations are negligible Figure 5.4 shows how some assump-tions can be correlated in the simulation For instance, the change in volatility in year 4 of the analysis is assumed to be correlated to the

volatility in the past three years Here we assume that the risk in the firm’s stock is autocorrelated.8Other examples may include a negative correla-tion between stock prices and forfeiture rates, and so forth

Rather than randomly deciding on the correct number of trials to run

in the simulation, statistical significance and precision control are set up to run the required number of trials automatically Figure 5.5 shows that a 99.9 percent statistical confidence on a $0.01 error precision control is

FIGURE 5.2 Distributional-fitting using historical, comparable, or forecast data.

Historical Data on Suboptimal Exercise Behavior

TABLE 5.2 Single-Point Result Using a Customized Binomial Lattice

Stock Price $100 Forfeiture Rate Blackout Dates