Pricing by arbitrage means that an asset is priced with the unique arbitrage-free price. Pricing by arbitrage can be applied in two different settings: (1) We can price by arbitrage linear securities, like forwards and futures, in any markets. (2) We can price by arbitrage nonlinear securities, like options, in complete markets. We discuss both of these cases in this chapter.
If two assets have the same terminal value with probability one, then the assets should have the same price. Otherwise, we could obtain a risk-free profit by selling the more expensive asset and by buying the cheaper asset. Linear assets, like futures, can be defined as a linear function of the underlying assets. Thus, they can be replicated, and their price is the initial value of the replicating portfolio. Nonlinear assets, like options, can be replicated only under restrictive assumptions on the markets (under the assumption of complete markets). However, the restrictive assumptions are often not too far away from the real properties of the markets.
The concepts of an arbitrage-free market and a complete market were studied in Chapter 13. The results of Chapter 13 imply that if the market is arbitrage-free and complete, then there is only one arbitrage-free price. In fact, Theorem 13.2 states that the arbitrage-free prices are obtained as expected values with respect to the equivalent martingale measures, and Theorem 13.3 (the second fundamental theorem of asset pricing) states that if the market is arbitrage-free and complete, then there is only one equivalent martingale measure.
Section 14.1 discusses pricing of futures, the put–call parity, and the American call options.
Section 14.2 studies binary models. In these models the price of a stock can at any time move only one step up, or one step down. The binary models are complete models, thus a derivative has a unique arbitrage-free price. These prices can be easily computed. The prices are called the Cox–Ross–Rubinstein prices, and they were introduced in Cox et al. (1979).
Section 14.2.3 studies asymptotics of multiperiod binary models, as the number of periods increases, and the length of the periods decreases. A multiperiod binary model converges to a log-normal model. In the log-normal model the stock price follows a geometric Brownian motion. The geometric Brownian motion is a market model that does not exactly describe the actual markets but it can be rather close to the actual markets. The Black–Scholes prices are obtained as the limits of the Cox–Ross–Rubinstein prices. The derivation of the Black–Scholes prices from the Cox–Ross–Rubinstein prices is both elegant and it leads to efficient numerical recipes, although there are many other derivations of the Black–Scholes price (see Section 14.3.3).
Section 14.3 studies the properties of the Black–Scholes prices. Section 14.4 studies Black–Scholes hedging. Section 14.5 studies combining the Black–Scholes hedging with time changing volatility estimates. This method of option pricing provides a benchmark against which we can evaluate competing pricing and hedging methods.
Futures can be priced by arbitrage regardless whether the market is complete or not. The payoff of a futures contract is a linear function of the underlying, and thus a futures contract can be replicated even in an incomplete market. The put–call parity provides a related example of pricing by arbitrage, which works whether the market is complete or not: the payoff of a put subtracted from the payoff of a call is a linear function of the underlying.
We consider pricing of stock futures, currency forwards, forward zero-coupon bonds, and forward rate agreements.
A futures contract on a stock is made at time . The contract specifies that the buyer of the contract has to buy the stock at a later time with price . We assume that the stock does not pay dividends during the time period from to . Let us denote with the price of the stock at time . The value of the futures contract at the expiration time is
because the buyer of the futures contract gives away and receives .
The price of a zero-coupon bond is taken as , where , is the annualized risk-free rate, and is the time between and in fractions of a year. This is the method of continuous compounding.1
What is the fair value of the futures contract for ? We may replicate the futures contract buy buying the stock and borrowing the amount . At time the value of this portfolio is
with probability one. At time the value of this portfolio is with probability one. Thus,
for , by the law of one price, to exclude arbitrage.
When an investor enters a futures contract in a futures exchange, this does not imply any cash flows, although the exchange requires from the investor a liquid collateral in order to secure a possible future payment. The fair forward price is called such value of that makes the value of the futures contract zero. To get we need that
The future prices that are quoted in a futures exchange are the values (which are determined by the supply and demand). Numbers are called futures prices or forward prices.
A buyer of a stock future or a stock index future saves the carrying costs but looses the possible stock dividends. When the annualized dividend rate is known to be , then the fair future price is
A currency forward is made at time . We denote with the exchange rate USD/euro at time . Let be an exchange rate. The contract stipulates that the buyer will buy US dollars with euros at time , using the USD/euro exchange rate . The value of the contract for the buyer at time is
Indeed, the buyer uses euros to buy dollars. Then the buyer exchanges the dollars to euros to obtain euros. The profit of the buyer is .
What is the price of the currency forward for ? We can replicate the currency forward using zero-coupon bonds. Let be the USD price of the US zero-coupon bond and let be the euro price of an European zero-coupon bond. The zero-coupon bonds are such that USD and . Let us consider the portfolio with units of US zero-coupon bonds and with units of European zero-coupon bonds. The value of the portfolio at time is
We choose
The value of the portfolio a time is
Since (14.3) and (14.4) are equal, the portfolio replicates the currency forward, and the price of the currency forward is equal to at time .
The price is in euros
To make we need to choose the exchange rate
A time two parties make an agreement to exchange at a future time a zero-coupon bond whose maturity is at , with a cash payment . At time only the agreement is made, and at time the cash payment is made and the zero-coupon bond is received. This is called a forward zero-coupon bond. Forward zero-coupon bonds are considered in Section 18.2.1.
Let be the price of the forward zero-coupon bond. Let and be the prices of zero-coupon bonds with maturities and . Then,
Consider a portfolio with one unit of and short of units of . The portfolio has value
for . Thus, and we have replicated the forward zero-coupon bond. Thus, at time the prices are equal:
To make , we need to choose the cash payment as
A forward rate agreement allows to change a floating Libor rate against a fixed rate. The contract is made at time . The reset time of the Libor is . At this time the Libor rate is settled to be , where is the maturity of the Libor. The agreement stipulates that the buyer of the contract pays the payment with the fixed rate , and receives the payment with the Libor rate . Thus, the payoff for the buyer at time is
where is the notional, and is the time between and . Forward rate agreements are considered in Section 18.2.2.
Let and be zero-coupon bonds. The forward rate agreement can be replicated by the trading strategy
where is the strategy of buying and reinvesting the payoff at time with the prevailing -period Libor rate. Thus, the payoff of at time is equal to . Thus, the trading strategy gives the payoff (14.6). The present value of the forward rate agreement is equal to
To make , we need to choose
Backwardation refers to the price relation where the spot price is higher than the forward price and contango refers to the case where the spot price is lower than the forward price:
At the expiration, the spot price and the futures price should be equal (to prevent arbitrage). Thus, the contango relationship means that distant delivery months trade at a greater price than near-term delivery months (the term structure is upward-sloping). Depending on the type of the underlying, either the contango or the backwardation is typical.
In the case of stock futures, or stock index futures, the theoretically fair price of a futures contract is greater than the spot price, according to (14.2). Thus, contango is typical for stock futures. However, futures prices can differ from the theoretically fair price, and thus both contango and backwardation are possible. In the case of stock futures we can change the terminology in the following way:
According to this terminology we do not compare futures prices to the spot prices but to the theoretically fair prices.2
When the underlying is a commodity, then there are borrowing costs for the seller of the futures contract, similarly as in the case of stock futures. In addition, there are costs for storing the commodity, for the seller of the futures contract. Thus, contango is typical for futures on commodities, and contango is typically more profound than for stock futures.
The buyer of a treasury bond future saves carrying costs (determined by the short-term rate) but looses the interest received by the bond owners (determined by the long-term rate). If the short-term rate is significantly less than the long-term rate, then the distant delivery months may trade at a lesser price than the near-term delivery months: this is backwardation.
The price of a put can always be expressed in terms of the price of a call, and conversely. We have the put–call parity:
where is the common strike price of the call and put, is the annualized interest rate, is the expiration time, and is the time from to in fractions of a year. It is clear that at the expiration we have . The put–call parity extends this relation for times before the expiration time . Thus, we do not need to know fair values for and in order to have an expression for their difference.3
Consider portfolio obtained by buying the call and writing the put:
At the expiration, we have . Indeed,
Let be the forward contract to buy stock at time with price . This forward contract was valued in (14.1), where we showed that
The replication was obtained by buying the stock and borrowing the amount . At the expiration, we have . Since with probability one, we have
for all times before , to exclude arbitrage. This is the statement (14.8) that we wanted to prove.
We have bounds for the call price :
The upper bound is obvious, since the right to buy a stock must be less valuable than the stock itself. To prove the lower bound, we can note that is obvious, since the right to buy a stock involves no obligations.4 The put–call parity and the fact that gives
We have proved the lower bound.5
The lower bound in (14.9) implies that
for . This follows because when . The inequality (14.10) leads to the definition of the time value of the option; see (2.6). The lower bound says that the value of the option is larger than the intrinsic value . The difference of the price of the option and its intrinsic value is called the time value of the European option. For an European put option the lower bound fails unless . As a consequence of the put–call parity, the time-value of a put option whose intrinsic value is large (the option is in the money), is usually negative.
An American call option has the same price as the corresponding European call option, when the stock does not pay dividends. On the other hand, an American put option has a different price than the corresponding European put option.
The put–call parity was derived for European options. However, this parity can be used to show that
when the stock does not pay dividends, where is the price of an American call option and is the price of the corresponding European call option. We know that , because an American option has more rights than the corresponding European option. The lower bound in (14.9) implies
where is the cash flow generated by the exercise of the American call option. Since , an early exercise is always suboptimal and one should sell the American call option and not exercise it. Since it is not optimal to exercise the option, the possibility for the early exercise is not worth of anything, and the American call option has the same value as the European call option.6
However, an American put option is in general worth more than the corresponding European put option. For the American options we have
The difference between the put and call options comes from the fact that the value of a put option increases as the price of the stock decreases. When the value of the stock decreases, then the absolute price changes become smaller. We can reach the point where the stock takes so small values that further decreases in the stock price would not give a better rate of return to the put option than the risk-free rate. At that point it is better to exercise the option and invest in the risk-free rate. With calls the situation reverses because as the stock price increases, absolute price changes increase. Pricing of American put options in the multiperiod binary model is discussed in Section 14.2.4.
We consider pricing and hedging of derivatives in binary models. In a binary model the stock can at any given time go only one step higher or one step lower. Section 14.2.1 studies one-period binary models. Section 14.2.2 uses one-period binary models as building blocks for multiperiod binary models. A multiperiod binary model approximates the Black–Scholes model, as is shown in Section 14.2.3.
First, we describe the one-period binary model. Second, we find the fair price and the optimal hedging coefficient. Third, we find the equivalent martingale measure. Fourth, we provide additional pricing and hedging formulas.
In the single period model, there are time points and . The market consist of stock , bond , and contingent claim .
The bond takes value 1 at and value at :
where is the risk-free rate.
The initial value of the stock is . The stock can take only two values at time 1. At the value is with probability and the value is with probability :
where . We assume that
Note that if , then it would not be rational to invest in the bond, and if , then it would not be rational to invest in the stock. The “rationality” can be formalized as the absence of arbitrage.
The contingent claim takes two possible values and at :
For example, in the case of a call option we have for some strike price . Then and .
We want to find a fair value for the derivative at . Also, we want to find the optimal hedging coefficient , which is used to hedge the position of the writer of the option.
We replicate the contingent claim with a portfolio
The portfolio consists of units of the bond and units of the stock. The portfolio takes values
To obtain we need to choose and so that
The first equation leads to
Inserting this value of to the second equation gives
The law of one price implies that the arbitrage-free price of the contingent claim equals the value of the replicating portfolio at time : , and thus
where and are defined in (14.13) and (14.14).
The number of bonds can be written as . Since , we can write
We can interpret the replicating portfolio from the point of view of the writer of the option as the portfolio where the writer receives the option premium , invests in the bank account, borrows the amount from the bank, and invests in the stock.
Price can be written as the expectation with respect to the equivalent martingale measure. We have
where
and means the expectation with respect to the probability measure with
Probability measure is called a risk-neutral measure because
and it is called a martingale measure because
Note that condition (14.12) guarantees that , so that is equivalent to .
It is of interest that the price can be written as
where
Indeed, similarly to (14.13), we get
which can be combined with (14.13) to get
Combining this formula for with the formula shows (14.19). In (14.19), we have written the derivative price as a discounted expectation of the derivative price, with an additional correction term.
The hedging coefficient can be written as
To derive (14.20), note that
Thus, (14.20) is equal to (14.14). We can also write
where is the gross return. This way of writing the hedging coefficient appears in (16.10), where quadratic hedging is considered.
We start with a description of the multiperiod binary model, and then proceed to the pricing formulas, derive the equivalent martingale measure, and give hedging formulas.
The market consists of stock , bond , and contingent claim . We define the discrete time processes
Note that in Section 13.2 we denoted the time steps of the discrete time markets by . We have changed the notation, because we construct a time series that approximates the geometric Brownian motion on . The approximation is done by dividing interval to periods of equal lengths and letting .
The bond takes value at step , where is the annualized interest rate, and is the time between two periods in fractions of a year.
At step , the stock can take values
where
The stock price is a random variable with
where ,
The stochastic process of stock prices can be described by a recombining binary tree, as in Figure 14.1, where . At step , the stock takes value (in the figure ). If the value of the stock at time is , then at step the stock can take values and , so that
The probabilities of the up and down movements are and :
The derivative can take at the expiration values , . The random variable takes the value , when . For example, when the contingent claim is the call option with , where is the strike price, then . We have
We want to find the arbitrage-free price of the derivative at time .
The evolution of the stock price in a multiperiod binary model has been described using a recombining binary tree. The price in a multiperiod binary model can be found by backward induction. We know the price at the expiration, when . We can use the single period model to calculate the price at step , and go backwards step by step to obtain the price at step . The price of the derivative is calculated using the backwards induction with the following steps.
We have described a recursive algorithm for the computation of the price, but we can also obtain the following explicit expression for the price.
The arbitrage-free price of the derivative is obtained not only at step , but the price is obtained at all steps . In fact, the price of the derivative, under the condition that the stock price at step is , is given by the formula
Let us define probability measure by
where . Measure is obtained from the physical measure , defined in (14.24), by replacing the probability of an up-movement by probability . The price in (14.27) can be written as the expectation
where is the random variable taking value , when takes value . The measure is called the equivalent martingale measure, or the risk-neutral measure.
The price of the derivative given in (14.28), under the condition that the stock price at step is , can be written as
Accordingly, we can define random variables
Defining
we obtain a more elegant formula
The following theorem gives the hedging coefficients of the replicating portfolio. The replication means that the wealth process of the trading strategy obtains the value of the derivative with probability one, or equivalently, the value process of the trading strategy obtains the discounted value of the derivative with probability one. The value process of a self-financing trading strategy is given in (13.8).
There are other ways to write the hedging coefficient. We have from the formula (14.21) of the one-period model that
where , , and . Here and mean conditional variance and covariance, conditional on , and under probability measure , which is the equivalent martingale measure.
We can also write the hedging coefficient as
Let us prove (14.33). We have from (14.31) that
Thus,
We have shown (14.33).
An additional formula for the hedging coefficient is
In fact, . Thus,
We start with the asymptotic normality of the logarithmic returns in the multiperiod binary model, then show the convergence of the arbitrage-free prices in the multiperiod binary model to the Black–Scholes prices, and finally show the convergence of the hedging coefficients in the multiperiod binary model to the Black–Scholes hedging coefficients.
Let
We consider asymptotics when . We choose the up and down factors as
where . We choose the probabilities of the up and down movements as
where . With these choices the logarithmic return converges in distribution to the normal distribution , where . Note that when we choose and as in (14.35), then the probability of the up movement in the risk-neutral distribution is
We show that the distribution of the stock price in the multiperiod binary model converges in distribution to a log-normal distribution, as the number of steps increases.
In the -period binary model the stock price , , can be written as
where are such i.i.d. random variables that when is a result of an up-movement and when is a result of a down-movement, so that
It holds that
as , where
We show a slightly more general result: For
as , where is such that , as . In particular, we can choose . Now (14.38) follows as a special case when we choose and .
We denote below . Let us denote . We can write
where
with . Now it holds that . We have that
as . Because , the claim (14.39) follows from items 1–3.
To prove item 1, we note that
Thus,
By the choice of , it holds as , and thus
as . Thus, item 1 follows by the central limit theorem.
To prove item 2, we use the fact that , which implies that as . Thus, the weak law of the large numbers implies item 2.
To prove item 3, we note that
Item 2 implies that and since , we have that , so that .
The arbitrage-free prices (14.27) in the multiperiod binomial model are called the Cox–Ross–Rubinstein prices. We show that the Cox–Ross–Rubinstein put and call prices converge to the Black–Scholes put and call prices, as . This is done in two steps. First, we show that the put and call prices in the multiperiod binary model converge to the expected values of the option pay-offs, when the expectation is taken with respect to a log-normal distribution. Second, we calculate closed-form expressions for the expected values.
A fundamental theorem about weak convergence states that if
then
as ; see Billingsley (2005, Theorem 25.8, p. 335).
Let be such function that Then the option payoff can be written as
If function is continuous and bounded, then (14.40) and (14.41) imply that
where the distribution of is defined by
and we used the fact that , because .
The fundamental theorem about weak convergence applies for any sequence converging weakly. However, in our case we are interested in the special case of the convergence of a binomial distribution toward a log-normal distribution. In this special case Föllmer and Schied (2002, Proposition 5.39, p. 265) notes that the convergence of expectations can be proved also when we relax the condition of the boundedness. Let be measurable, almost everywhere continuous, and
Then,
The payoff function of a put option is
where is the strike price. Function , , is bounded and continuous. Thus, the Cox–Ross–Rubinstein price of an European put option converges to the expectation:
The right hand side is equal to the Black–Scholes put price, as shown in (14.54).
The payoff function of a call option is
where is the strike price. Function , , is continuous but not bounded. Thus, the convergence of the Cox–Ross–Rubinstein price to the expected value cannot be inferred similarly as in the case of put options. However, we can apply the put-call parity in (14.8) to conclude that the Cox–Ross–Rubinstein prices have to satisfy
where is the Cox–Ross–Rubinstein call price, and the put price. Since we have shown (14.45), it holds that
The right hand side is equal to the Black–Scholes call price, as shown in (14.52).
We have proved in (14.45) and (14.46) that the arbitrage-free put and call prices in the multiperiod binary model approach the expected values of the option payoffs, when the expectations are with respect to the equivalent martingale measure (the risk neutral log-normal model). Thus, we want to calculate the expected values of the put and call payoffs. The expected values are the Black–Scholes prices, which we denote
and
where the expectations are taken with respect to the distribution of defined by
where and
Note that we have discussed the log-normal distribution in (3.50).
The density of the standard normal distribution is , where . Then,
where
By writing we have
Thus,
where is the distribution function of the standard normal distribution. Since ,
where
because and . This leads to the call price
Similarly,
which leads to the put price
We have calculated explicit expressions for the call and put prices at time , when the stock price is . The Black–Scholes call and put prices at time , when the stock price is , are given by
and
where the expectations are taken with respect to the distribution of defined by
where and . The corresponding explicit expressions are given in (14.58) and (14.59), and the Black–Scholes prices are discussed in Section 14.3.1.
We show that the hedging coefficients of puts and calls in the multiperiod binary model converge to the Black–Scholes hedging coefficients. The Black–Scholes hedging coefficients are called deltas, and they are obtained by differentiating the Black–Scholes prices with respect to the stock price.
Let the initial time and the initial stock price be
Let be either the Black–Scholes call price or the put price at time , when the stock price at time is , and the expiration is at time . These prices are written as expectations in (14.55) and (14.56), and in a more explicit form in (14.58) and (14.59).
Let us consider step of the multiperiod binary model. Let be one of the possible stock prices at step , where . These prices were defined in (14.23). Let be such that
as . In particular, we can choose , where is the largest integer . Choose
Then
as .
The hedging coefficient of the multiperiod binary model was written in (14.32) as
where and are the two possible prices of the derivative at step , when the value of the stock at step is .7 Now it holds that
where is the derivative of the Black–Scholes price with respect to stock price.
To show (14.57), note that from (14.46) and (14.45), and the continuity of the functions we have that
where means that . Thus,
Also, , , and , as .
Figure 14.2 illustrates the convergence of (a) the ratio of the Cox–Ross–Rubinstein call price to the Black–Scholes call price, and (b) the ratio of the Cox–Ross–Rubinstein call hedging coefficient to the Black–Scholes call hedging coefficient. The ratios are plotted as a function of the number of the steps in the multiperiod binary model, where . The moneyness is 0.9 (green), 0.95 (black), and 1 (red). The annualized volatility is , the interest rate is , and the time to maturity is 1 month: . We see that the convergence of the at-the-money options is faster than the convergence of the out-of-the-money options.
We have proved the weak convergence at one point: Let and . Then
where is the price of the stock at step in the multiperiod binomial model, and is the price of the stock at time in the geometric Brownian motion model. If the option payoff is , then the price in the multiperiod binomial model converges to the price of the option whose payoff is in the geometric Brownian motion model. Asian options and knock-out options depend on values of stocks in more than one point.
In the case of Asian options, the option payoffs can be written as
where are steps, and are fixed time points. For example, in the case of an Asian call option
The corresponding prices converge for a suitable , when
The payoffs of knock-out options depend on the trajectory of the prices through
Let us divide the time interval into subintervals. Denote the boundaries of the intervals by
Points fill the interval asymptotically. We can define a continuous time process , , by linearly interpolating : define
when . Geometric Brownian motion is defined in (5.62). We can show that process converges weakly to the geometric Brownian motion
where is the standard Brownian motion. The weak convergence
as , happens in the metric space of the continuous functions on . The prices of the options whose payoffs are
converge for suitable .
An American put option has a different price than the corresponding European put option, and the price of an American put option does not have a closed-form expression in the multiperiod binary model. However, an American call option has the same price as the corresponding European call option, when the stock does not pay dividends (see Section 14.1.3). Thus, the American call options can be priced similarly as the European call options using the Black–Scholes prices or the recombining binary trees.
The American put options have to be priced by taking into account the possibility of an early exercise. We can use the recombining binary tree to price the American put options. At every node of the tree we consider whether it is better to exercise or to keep the option for a future exercise. We are not able to obtain a closed-form formula for the price of the American put options, but we obtain an algorithm for the computation of the price. First, the single period binary model is studied. Second, the multiperiod binary model leads to the final algorithm.
In the one-period binary model, the American put option can be exercised at time or at time . Let us denote with the value of the American put option at time and let us denote with the value of the European put option at time . An arbitrage argument shows that
The value of an European put option can be obtained from (14.17) as
where
with
The price of an American put option is determined in the -step binomial model by recursion. Remember that in the -step binomial model the possible prices at step , , are
with
The recursive steps are the following.
.
First we describe the properties of Black–Scholes call and put prices, second we discuss implied volatility, third we describe various ways to derive the Black–Scholes prices, and fourth we give Black–Scholes formulas for options on forwards, for fixed income options, and for currency options.
The Black–Scholes price of the call option at time , with strike price , and with the maturity date , is equal to
where is the stock price at time , is the annualized risk-free rate,
and is the distribution function of the standard Gaussian distribution. The time to expiration is expressed in fractions of a year. The put price is equal to
Note that it can be convenient to write
The Black–Scholes price is derived under the assumption of a log-normal distribution of the stock price: It is assumed that at time
where , is the drift, and is the volatility. Note that under the risk-neutral measure . The volatility is the only unknown parameter that need to be estimated, since does not appear in the price formula.
For the application of the Black–Scholes formula the time is taken as the time in fractions of year. For example, when the time to expiration is 20 trading days, then . Alternatively, when time to expiration is 20 calendar days, then .
The risk-free rate is expressed as the annualized rate.
The only unknown parameter has to be estimated. Let be an equally spaced sample of stock prices and let us denote for . We assume that , are i.i.d. , so that the stock prices satisfy (3.50). We can estimate with the sample variance
where . Then an estimator of is
For example, if we sample stock prices daily, then and .8 If we sample stock prices monthly, then and The normalized sample standard deviation in (14.60) is called the annualized sample standard deviation.
We study the qualitative behavior of the Black–Scholes prices as a function of five parameters , , , , and .
The prices of calls and puts increase as increases. We have
and
which are the bounds derived from the put–call parity in (14.9). The prices of calls and puts increase as the time to maturity increases. The price of a call increases as increases and the price of a call decreases as increases, but for puts the relations reverse. The price of a call increases as the interest rate increases but the price of a put decreases as the interest rate increases.
Figure 14.3 shows Black–Scholes prices for calls and puts as a function of the call moneyness . The call prices increase and the put prices decrease as a function of moneyness. Panel (a) shows the cases of annualized volatility (black), (red), and (green). The time to maturity is 20 trading days and the interest rate is . Panel (b) shows the cases of interest rates (black), (red), and (green). The time to maturity is 20 trading days and the annualized volatility is . We see from panel (a) that increasing volatility increases the prices, both for call and puts. The effect of increasing time to maturity is similar. We see from panel (b) that for calls increasing interest rates increases prices, but for puts increasing interest rates decreases prices.
Figure 14.4 shows that the call and put prices are not symmetric. The ratios of call prices to put prices are shown as a function of the call moneyness : We show the functions
where is the current stock price. The strike prices for the calls take values , and the corresponding strike prices for the puts are . Panel (a) shows the cases of annualized volatility (black), (red), and (green). The interest rate is . Panel (b) shows the cases of interest rates (black), (red), and (green). The annualized volatility is . The time to maturity is 20 trading days in both panels. We see that the call prices are higher than the put prices for out-of-the-money options. This is related to the asymmetry of the log-normal distribution (see Figure 3.11). For at-the-money options the difference between the call and the put prices is not large. Increasing the volatility makes the ratios of the call and put prices closer to one, and decreasing the interest rate makes the ratios of the call and put prices closer to one.
To apply Black–Scholes prices we need to estimate the volatility. We study how the Black–Scholes prices change when the volatility estimate changes. We apply the data of S&P 500 daily prices, described in Section 2.4.1.
Figure 14.5 shows time series of Black–Scholes call prices. The volatility is equal to the sequential annualized sample standard deviation. Panel (a) shows call prices with moneyness and panel (b) shows call prices with moneyness . The time to expiration is either 20 trading days (black curves) or 30 trading days (red curves). The risk-free rates are deduced from 1-month Treasury bill rates. We can see that the time series of Black–Scholes prices follow closely to the time series of sequential standard deviations in Figure 7.5.
Figure 14.6 studies Black–Scholes prices when the volatility is equal to the sequentially estimated GARCH() volatility. Panel (a) shows time series of call prices with moneyness (black) and (green). The time to expiration is 20 trading days. Panel (b) shows kernel density estimates of the distributions of the prices. The horizontal and the vertical lines show the Black–Scholes prices when the volatility is the annualized sample standard deviation computed from the complete sample. The risk-free rate is zero.
The greeks are defined by differentiating the option price with respect to the price of the underlying, the time to the expiration, the interest rate, or the volatility. In this section, we denote the Black–Scholes call and put prices by
where is the current time, is the current price of the underlying, is the strike price, is the time of the expiration, is the volatility, and is the interest rate. Sometimes we leave out some of the arguments and denote, for example, .
The delta is the derivative of the price function with respect to the underlying. The delta is the hedging coefficient in the Black–Scholes hedging, as discussed in Section 14.4. The call and put delta are
We have that
In fact, if increases, then the price of the call increases and the price of the put decreases: and . The absolute value of the change in the value of a call or a put cannot exceed the absolute value of the change in the underlying: and .
The call delta is equal to
and the put delta equal to
Let us calculate the call delta. The delta of a call is given in (14.62) because
and finally9
Figure 14.7 shows the call delta as a function of moneyness . In panel (a) the annualized volatility is 10% (black) and 20% (red) with time to expiration 1 month. In panel (b) the time to expiration is 1 month (black) and 3 months (red) with volatility 10%. The solid lines show the case of interest rate and the dashed lines show the case of interest rates . The more interesting part is the moneyness . In this region the deltas are in both panels. Panel (a) shows that when moneyness is , then a larger volatility leads to a larger delta. Panel (b) shows that increasing time to maturity has a similar qualitative effect as increasing volatility.
Figure 14.8 shows a time series of Black–Scholes deltas using the data of S&P 500 daily prices, described in Section 2.4.1. The daily prices are used to create a time series with the sampling frequency of 20 and 30 trading days. Panel (a) shows call prices with moneyness and panel (b) shows call prices when moneyness . The time to expiration is either 20 trading days (black curves) or 30 trading days (red curves). The volatility is equal to the sequential sample standard deviation. The risk-free rates are deduced from 1-month Treasury bill rates. The corresponding time series of Black–Scholes prices is given in Figure 14.5.
The gamma is the second derivative of the price function with respect to the underlying:
The price functions are convex with respect to and thus
The call gamma and the put gamma are given by
Figure 14.9 shows the call gamma as a function of moneyness . In panel (a) the annualized volatility is 10% (black) and 20% (red) with time to expiration 1 month. In panel (b) the time to expiration is 1 month (black) and 3 months (red) with volatility 10%. The solid lines show the case of interest rate and the dashed lines show the case of interest rates .
The theta is the derivative of the price function with respect to time:
As increases the value of the option decreases (everything else being equal) and thus
The call theta is equal to
and the put theta is equal to
Figure 14.10 shows the call theta as a function of moneyness . In panel (a) the annualized volatility is 10% (black) and 20% (red) with time to expiration 1 month. In panel (b) the time to expiration is 1 month (black) and 3 months (red) with volatility 10%. The solid lines show the case of interest rate and the dashed lines show the case of interest rates .
The vega is the derivative of the price function with respect to volatility:
The call vega and the put vega are equal to
Figure 14.11 shows the call vega as a function of moneyness . In panel (a) the annualized volatility is 10% (black) and 20% (red) with time to expiration 1 month. In panel (b) the time to expiration is 1 month (black) and 3 months (red) with volatility 10%. The solid lines show the case of interest rate and the dashed lines show the case of interest rates .
The rho is the derivative of the price function with respect to interest rate:
The call rho is equal to
and the put rho is equal to
Figure 14.12 shows the call rho as a function of moneyness . In panel (a) the annualized volatility is 10% (black) and 20% (red) with time to expiration 1 month. In panel (b) the time to expiration is 1 month (black) and 3 months (red) with volatility 10%. The solid lines show the case of interest rate and the dashed lines show the case of interest rates .
Implied volatilities can be derived both from call prices or from put prices. Let be a Black–Scholes call price. The mapping
is bijective and can be inverted. Let us denote by the inverse of mapping (14.64). When we observe a market price for a call option, then
is the implied volatility of the option. The implied volatility of a put option can be defined similarly.
It is helpful to quote option prices using implied volatilities. Option prices and implied volatilities are in a bijective correspondence, but it is easier to compare the prices of options with different maturities and strike prices using the implied volatilities than using the market prices. This is analogous to expressing the bond prices with annualized rates instead of quoted prices.10 The implied volatilities can be used to quote prices even when we do not think that the Black–Scholes prices are fair prices, similarly as the bond rates can be defined using various conventions.
If the Black–Scholes model describes the true distribution of the asset prices, and if the market prices coincide with the Black–Scholes prices, then the implied volatilities of options with different strike prices and with different maturities are all equal. However, in practice the implied volatilities are different for the options with different strike prices and with different maturities.
The volatility surface gives for each strike price and for each maturity the corresponding implied volatility. Let be the market prices of call (put) options with strike prices and expiration dates , where , . The options are otherwise similar. The volatility surface is the function
where is the time to the expiration.
The volatility surface is typically not a constant function. Instead, for a fixed maturity , function
is typically u-shaped (smile) or skew (one sided smile). Instead of the strike one may take as the argument the moneyness (or ), or the delta of the options.
Options written on equity indices yield often skews. This might be due to the fact that a crash in stock markets leads to increased volatility, whereas a rise in stock markets is not usually involved with an increased volatility. Options on various interest rates yield more monotonous one sided smiles than the equity indices.
Currency markets yield often symmetric smiles. Currency markets are more symmetric than stock markets because a big movement in either direction results in an increased volatility (it is always a crash for one of the currencies).
The smile tends to flatten out with maturity. This might be due to the better Gaussian approximation when the maturity is longer.
Out-of-the-money options can be priced in the following way. (1) Find the implied volatility of at-the-money options. (2) Adjust the implied volatility (using experience) to get a new volatility. (3) Calculate the price of the out-of-the-money option using the new volatility.
The VIX index of CBOE (Chicago Board Options Exchange) uses prices of options to derive the volatility that is expected by the markets. Section 6.3.1 contains a discussion of the VIX index and Figure 6.5 shows a time series of the VIX index. Let us derive the formula of the VIX index.
It is assumed that the stock price follows a geometric Brownian motion, as defined in (5.62). That is,
where is the standard Brownian motion, , and . Under the equivalent martingale measure , where is the yearly risk-free interest rate. Thus, for the equivalent martingale measure we have
We can solve for to get
A Taylor expansion gives
where . Thus,
Theorem 13.2 implies that when the expectation is taken with respect to a risk-neutral measure, then
where and are the arbitrage-free prices of the call and the put. Under the risk-neutral measure
where is the futures price, as given in (14.2). Thus,
We arrive at the variance formula
The variance formula (14.65) was derived in Demeterfi et al. (1999). CBOE uses the approximation
CBOE:s VIX index is defined as11
where is the midpoint of the bid-ask spread for the option, and . The calculation is done for two expiration dates, and the final index value is a weighted average of these. Note that the put–call parity (14.8) gives the equality .
We have derived the Black–Scholes prices as the limits of the prices in the multiperiod binary model. Now we describe shortly the martingale derivation of the Black–Scholes prices, derivation of the prices using the Black–Scholes differential equation, and the derivation of the prices using the put–call parity.
The second fundamental theorem of asset pricing says that an arbitrage-free market model is complete if and only if there exists exactly one equivalent martingale measure. We have stated this theorem for the discrete time model in Theorem 13.3.
The Black–Scholes prices can be derived as a corollary of the second fundamental theorem of asset pricing: If the Black–Scholes market model is arbitrage-free and complete, then the Black–Scholes prices are the discounted expectations with respect to the equivalent martingale measure.
Shiryaev (1999, p. 710) states in a continuous time framework that if there exists a unique equivalent martingale measure, then the unique arbitrage-free price of the option is the discounted expected value of the payoff with respect to the unique equivalent martingale measure.
We give a sketch of some elements of the derivation of the Black–Scholes prices directly from the second fundamental theorem of asset pricing. Details can be found in Shiryaev (1999, p. 739).
The Black–Scholes model assumes that the stock price follows the geometric Brownian motion, as defined in (5.62). The stock price satisfies
and the bank account satisfies
where .12 Thus,
The Girsanov's theorem was stated in (5.64). We apply Girsanov's theorem with the constant function . Then,
is a Brownian motion with respect to measure , defined by , where
Measure is the unique martingale measure that is equivalent to ; see Shiryaev (1999, p. 708). Thus, the price of the call option is
We need to find the distribution of under . From (14.68) we obtain that
Thus,
Black and Scholes (1973) and Merton (1973) derived the Black–Scholes price by solving a differential equation. The Black–Scholes partial differential equation is
where is the value of the option at time , is the theta of the option, is the delta of the option, is the gamma of the option, and . When is the value of a call option, then the solution is found under the boundary condition
The price of the option is . The differential equation is solved, for example, in Shiryaev (1999, p. 746).
The Black–Scholes partial differential equation can be derived heuristically in the following way. Itô's lemma (5.61) applied to the function gives
Assume that value of the option is replicated by the portfolio
that is,
The assumptions and imply
Equating (14.70) and (14.71) gives , that is,
This means that we want a perfect replication of . Choose , which is called delta hedging. This makes to disappear and leads to the differential equation (14.69). Delta hedging has removed all uncertainty, and we have obtained a perfect hedge.
We can derive the Black–Scholes prices for the calls and puts using the put–call parity given in Section 14.1.2. The basic idea is that if we are willing to assume that the prices are expectations with respect to a log-normal distribution, then the put–call parity implies that the log-normal distribution should be the risk-neutral log-normal distribution.
We assume that the distribution of the stock price is defined by (14.49). Let us denote by the price of the call option at time . We assume that the price of the call option is equal to
and the value of the put option is equal to
Thus, using (14.50) and (14.53),
because for all . The put–call parity (14.8) implies that we have to take
Inserting (14.72) to (14.50) and (14.53) leads to (14.58) and (14.59). This derivation was noted in Derman and Taleb (2005).
We give some examples of Black–Scholes prices. The examples include pricing functions of options on a forward, caplets, swaptions, options on a foreign currency, and barrier options.
Let the underlying be a futures contract with the maturity . Consider a call option with the expiration time . The payoff is , where is the strike price. The price of the call option is obtained by replacing in the Black–Scholes formula (14.58) by . This gives the price
where
The volatility is the volatility of the stock. This is called Black's formula, and it was introduced in Black (1976).
Why have we replaced with ? This is due to the fact that under the risk-neutral measure the stock price has the distribution
The pricing formula of the futures contract in (14.2) gives . Thus, the distribution of under the risk-neutral measure is
The distribution of is
where . The price of the call option is
and this expectation is calculated similarly as in (14.51).
The formula (14.73) holds when the option is subject to the stock type settlement. If the option is subject to the futures type settlement, then set in (14.73). The futures type settlement means that the gains and losses are realized daily, whereas in the stock type settlement the gain or loss is realized at the time of liquidation.
Caplets are discussed in Section 18.3.1. A caplet is a call option on the Libor rate . The payoff of a caplet at time is
where is the principal, and is the strike. We assume that under the risk-neutral measure
where and is the forward rate, defined in (18.14). Note that at time the forward Libor rate is equal to the spot Libor rate: . The Black's formula for the price of the caplet is
where
The caplet price can be written as the expectation
where the expectation is with respect to the risk-neutral distribution in (14.74). The expectation is calculated similarly as in (14.51). The caplet price is obtained from the Black–Scholes price of a call option on a stock when is replaced by and is replaced by .
Caps are defined in Section 18.3.2. Let be the time points for the caplets on the Libor rates , where . A cap is priced by
where are the prices of the caplets.
Swaptions are discussed in Section 18.3.3. Let be the current time, be the expiry time, and be the maturity time of the swaption. The payoff of an European call option on a swap is given in (18.27) as
where is the equilibrium swap rate, is the strike,
is the principal, and is a zero-coupon bond. We assume that under the risk-neutral measure
where . The Black's formula for the price of the swaption is
where
The swaption price can be written as the expectation
where the expectation is with respect to the risk-neutral distribution in (14.74). The expectation is calculated similarly as in (14.51). The caplet price is obtained from the Black–Scholes price of a call option on a stock when is replaced by and is replaced by .
Note that the simultaneous log-normality for all caplets and all swaptions is not consistent because a swap rate is a linear combination of forward rates and cannot be log-normal if the underlying forward rates are.
Let be the price of a foreign currency in the domestic currency units. The payoff of an European call option on a foreign currency is given by
where is the strike, and is the expiration time. According the Garman–Kohlhagen model, under the risk-neutral measure,
where . Then the price of the call option is
where
where is the risk-free rate in the foreign currency and is the risk-free rate in the domestic currency. The put price is
The price of the call option can be written as
where the expectation is with respect to the risk-neutral measure. The expectation is calculated similarly as in (14.51).
The call price on the exchange rate is obtained from the Black–Scholes price of a call option on a stock when is replaced by the exchange rate and is replaced by .
A down-and-out call on stock has the payoff
where is the strike price and is the barrier. We assume that the stock price has a log-normal distribution
under the risk-neutral measure, where . The price of the down-and-out call is
where is the Black–Scholes price of the vanilla call and
where
The hedging coefficients of calls and puts are given in (14.61). The Black–Scholes hedging coefficients are equal to the deltas of options. A delta is the derivative of the price of the option with respect to the price of the underlying: the call and put deltas are
This is shown in Section 14.2.3; see (14.57).13 For the Black–Scholes pricing functions the call delta and the put delta are shown in (14.62) and (14.63) to be equal to
and
In this section, our purpose is to illustrate how hedging can be used to approximately replicate options, and to study how hedging frequency, expected stock returns, and the volatility of stock returns affect the replication. The study is made using Black–Scholes hedging, since Black–Scholes hedging provides a benchmark for comparing various hedging methods.
We take the purpose of hedging to be to make the probability distribution of the hedging error of the writer of the option as concentrated around zero as possible. We consider S&P 500 options and estimate the distribution of the hedging error of the writer using the S&P 500 daily data of Section 2.4.1.
Section 14.4.1 reviews historical simulation for the estimation of the distribution of the hedging error. Section 14.4.2 studies the effect of hedging frequency to the distribution of the terminal wealth. Section 14.4.3 studies the effect of the strike price, Section 14.4.4 studies the effect of the mean return, and Section 14.4.5 studies the effect of the return volatility.
We have discussed the estimation of the hedging error using historical simulation in Section 13.3.1. Let us write the formulas for hedging error again, this time taking into account the varying hedging frequencies.
The hedging error of the writer of the option is obtained from (13.10) by the formula
where
when we take the risk-free rate . Here is the price of the option, is the terminal value of the option, are the hedging coefficients, and are the stock prices. In (14.78) the current time is denoted by , the time to expiration is days, and hedging is done daily. The hedging can be done with a lesser frequency. When hedging is done times during the period , then
where
Let us denote the time series of observed historical daily prices by . Let us denote the time to the expiration by
(The interpretation is that the last observation is made at time , and we are interested to write an option whose expiration time is .) We construct sequences of prices:
where
for . Each sequence has length and the initial prices are always . We estimate the distribution of the hedging error from the observations
where is computed from the prices . For the estimation of the density we use both the histogram estimator and the kernel density estimator, defined in Section 3.2.2.
We consider hedging of call options. The Black–Scholes prices are given in (14.58) and the hedging coefficients are given in (14.62). The volatility is estimated using the annualized sample standard deviation, computed from the complete data. We study the effect of volatility estimation to hedging in Section 14.5, where sequential (out-of-sample) volatility estimation is studied. In this section, we use in-sample volatility estimation.
More precisely, the th hedging error is
where is the Black–Scholes price from (14.58), computed with the stock price , time to expiration , and volatility . Here is the annualized sample standard deviation computed from return data
where
. Thus, the volatility is estimated in-sample. The hedging coefficient is the Black–Scholes delta from (14.62), computed with the stock price , time to expiration , and volatility . The terminal value is .
In this section, we illustrate how a change in the hedging frequency changes the distribution of the hedging error. We study hedging of S&P 500 call options, using daily data of Section 2.4.1. The time to expiration is 20 days. The volatility in the Black–Scholes formula is the non-sequential annualized sample standard deviation.
Figure 14.13 shows time series of hedging errors. Panel (a) shows the case when there is no hedging, so that . Panel (b) shows the case when the hedging is done daily. Moneyness is .
Figure 14.14 shows tails plots of the hedging errors. Panel (a) shows the left tail plots and panel (b) shows the right tail plots. We show cases of no hedging (red), hedging once (black), hedging twice (blue), and hedging 20 times (green).
Figure 14.15 shows (a) histograms of hedging errors and (b) kernel density estimates of the distribution of hedging errors. Panel (a) shows cases of no hedging (red) and daily hedging (green). Panel (b) shows additionally the cases of hedging once (black) and hedging twice (blue). Note that the red kernel density estimate is very inaccurate, because the underlying distribution is such that a large part of the probability mass is concentrated at one point.
Figure 14.16 shows tails plots of the hedging errors. Panel (a) shows the left tail plots and panel (b) shows the right tail plots. We show cases of no hedging (red), hedging once (black), hedging twice (blue), and hedging 20 times (green).
Figure 14.17 shows (a) histograms of hedging errors and (b) kernel density estimates of the distribution of hedging errors when the moneyness is . Panel (a) shows cases of no hedging (red) and daily hedging (green). Panel (b) shows additionally the cases of hedging once (black) and hedging twice (blue). The red kernel density estimate is very inaccurate, because the underlying distribution is such that a large part of the probability mass is concentrated at one point.
Figure 14.18 shows the estimated expected utility as a function of the hedging frequency. In panel (a) the moneyness is . In panel (b) . We apply the exponential utility function , where the risk-aversion parameter takes values (black), (red), and (blue). The expected utilities are estimated using the sample averages. Increasing hedging frequency clearly increases the expected utility. We can see that when the risk aversion is small, then the increase in the expected utility as a function of hedging frequency is much smaller than the increase when the risk aversion is large.
In this section, we illustrate how a change in the strike price changes the distribution of the hedging error. We study hedging of S&P 500 call options, using daily data of Section 2.4.1. The time to expiration is 20 days and hedging is done daily. The volatility in the Black–Scholes formula is the nonsequential annualized sample standard deviation.
Figure 14.19(a) shows tail plots of hedging errors and panel (b) shows kernel density estimates of the distribution of the hedging error. The strikes prices of calls are (red), (black), and (blue), when the stock price is . For in-the-money options the distributions have a larger spread than for out-of-the-money options. Also, the distributions for at-the-money options have a center that is located to the right from the centers for out-of-the-money options.
We study the effect of the mean return to the distribution of the hedging error. This is done by manipulating the S&P 500 data. We change observations so that the new net returns are
where are the observed net returns, is the sample mean of , and is a value for the expected return that we choose. The new prices are obtained from the net returns by
For the S&P 500 returns the annualized mean return is about . We try the annualized expected returns and . Thus, and .
Figure 14.20 shows histogram estimates of the distribution of the hedging error when the expected annualized return is . In panel (a) there is no hedging and in panel (b) there is daily hedging. That is, panel (a) shows a histogram made from realizations of the random variable , where , and is the Black–Scholes price. Panel (b) shows a histogram made from realizations of the random variable . The time to the expiration of the call option is days, and the strike price is with the initial stock price . Since is large the call option gives a profit to its owner with a large probability, as can be seen from panel (a). However, large does not change much the distribution of the hedging error when hedging is done daily, as can be seen from panel (b). In fact, the corresponding distribution of the hedging error when the expected return is moderate is shown in Figure 14.15.
Figure 14.21 shows the setting of Figure 14.20 when the annualized expected return is , instead of . Panel (a) shows the distribution of the hedging error of the writer when no hedging is done and panel (b) shows the hedging error when delta hedging is done daily. Since the expected return is negative, the writer of the call option gets a profit with a large probability, but this does not affect much the hedging error when the hedging is done daily, as can be seen from panel (b).
(1) When the drift is larger than the risk-free rate, then the Black–Scholes price is smaller than the expectation . The possibility of hedging makes smaller than . The expectation increases when the drift increases but the possibility of hedging makes the price independent of the drift. (2) When the drift is equal to the risk-free rate, then is close to , but hedging reduces the risk of the writer of the option, because it changes the wealth distribution of the writer of the option. (3) When the drift is negative, then is smaller than , and hedging reduces the expected profit of the writer of the option. However, the hedging reduces also the risk of the writer of the option, and thus hedging is reasonable even in the case of negative drift.
We study the effect of the return volatility to the distribution of the hedging error. We manipulate the S&P 500 data by changing observations so that the new net returns are
where are the observed net returns, is the sample standard deviation of , is the sample mean of , and is a value of our choice for the volatility. The new prices are obtained from the net returns by
For the S&P 500 returns the annualized sample standard deviation is about . We try the annualized standard deviation . Thus, .
Figure 14.22 shows histogram estimates of the distribution of the hedging error when the annualized volatility is . In panel (a) there is no hedging, and in panel (b) there is daily hedging. That is, panel (a) shows a histogram made from realizations of the random variable , where , and is the Black–Scholes price. Panel (b) shows a histogram made from realizations of the random variable . The time to the expiration is days, and the strike price is with the initial stock price . We see that the larger volatility makes the dispersion of the probability distribution of the hedging error larger.
We continue to study the distribution of the hedging error , as in Section 14.4. In this section, our aim is to study how the volatility estimation affects the distribution of the hedging error. The Black–Scholes prices and the hedging coefficients depend on the annualized volatility . We compare the performance of GARCH() and exponentially weighted moving averages for the estimation of . The performance of Black–Scholes hedging will be used as a benchmark.
We have discussed in Section 13.3.1 the estimation of the distribution of the hedging error using historical simulation. We write again the formulas for the hedging error and historical simulation, adapting to the current setting.
The hedging error of the writer of the option is obtained from (13.10) as
where
Here the risk-free rate is , is the price of the option, is the terminal value of the option, are the hedging coefficients, are the stock prices, the current time is denoted by 0, the time to expiration is days, and hedging is done daily.
We denote the time series of observed historical daily prices by . We construct sequences of prices:
where
for . Each sequence has length and the initial price in each sequence is . We estimate the distribution of the hedging error from the observations
where is computed from the prices .
More precisely, the th hedging error is
where is the Black–Scholes price from (14.58), computed with the stock price , time to expiration , and volatility . Here is the annualized volatility estimate computed using return data
where
Thus, the volatility estimation is done sequentially (out-of-sample). The hedging coefficient is the Black–Scholes delta from (14.62), computed with the stock price , time to expiration , and volatility . The terminal value is .
For the estimation of the density we use both the histogram estimator and the kernel density estimator, defined in Section 3.2.2.
In the following examples the time to expiration is trading days. We start pricing and hedging after 4 years of data has been collected. The risk-free rate is equal to zero. We consider S&P 500 call options and use the daily S&P 500 data, described in Section 2.4.1.
As a summary of the results, we can note that the GARCH() and the exponential moving average (with a suitable smoothing parameter) improve the distribution of the hedging error from the point of view of the writer of the option, when compared to the sequential sample standard deviation. GARCH() and the exponential moving average lead to a distribution whose left tail is lighter: with these volatility estimators the losses of the writer of the option are smaller. On the other hand, GARCH() and the exponential moving average lead to larger positive hedging errors. The positive hedging errors are gains for the writer of the option.
Figure 14.23 shows (a) the means of negative hedging errors and (b) the means of positive hedging errors as a function of the moneyness . The arithmetic means of positive and negative hedging errors are defined as
where the hedging errors are defined in (14.81). The hedging is done with sequentially computed sample standard deviation (red with “s”), GARCH() volatility (blue with “g”), the exponential moving average with (yellow with “1”), and with smoothing parameter (green with “2”). We see that the sample standard deviation is the best for the positive hedging errors and the worst for the negative hedging errors. The performance of GARCH() and the exponentially weighted moving average with are close to each other. The exponentially weighted moving average with has the performance between the sample standard deviation and the GARCH(). We can see that hedging errors are larger for the at-the-money options than for the out-of-the-money options.
Figure 14.24 shows (a) tail plots of hedging errors and (b) kernel density estimates of the distribution of the hedging error. The moneyness of call options is . The volatility is estimated by the sequentially computed standard deviation (red), by GARCH() (blue), and by the exponentially weighted moving average with (green). Tail plots are defined in Section 3.2.1 and the kernel density estimator is defined in Section 3.2.2. We apply the standard normal kernel function and the smoothing parameter is chosen by the normal reference rule.
Figure 14.25 shows (a) tail plots of hedging errors and (b) kernel density estimates of the hedging error. The volatility is estimated by the sequentially computed standard deviation (red), and by the exponentially weighted moving average with (purple), (green), (blue), and (black).
since for the risk-neutral measure .
where is the number of stocks that are bought at time to hedge the position until . The amount is borrowed with the risk-free rate at time . If the hedging gives a position without risk, we should have , which gives
when . This gives the instantaneous optimal hedging coefficient as