Published on: **Mar 4, 2016**

Source: www.slideshare.net

- 1. Pricing and Hedging of Japan Equity-linked Power Reverse Dual Note Using Least SquaresMonte Carlo SimulationFrank FungShaoyu HouNicholas LeeMayank RaoAbstractWe computed the price of a structured product, the Japan Equity-linked Power Reverse Dualnote (hereafter referred to as JRD), using a Monte Carlo approach with Least Squares toaccount for early exercise. We benchmarked the pricing model on a number of simplerderivatives with closed form solutions or highly accurate pricing from FD schemes. Thesensitivity of JRD with respect to various parameters and performance from different basisfunctions is investigated. We employed quasi-random sequences, latin hypercubes andBrownian bridges as Variance Reduction techniques. We also hedge the JRD using static anddynamic delta hedging.
- 2. NomenclatureFor illustrative purpose we consider an instance of the product with 8 months until maturity,2 coupon payments in total (including that at maturity) and monthly observations: τ tଵ TT: Date of maturityτ: Time interval between two consecutive observation datest ୬: The date of the payment of the nth coupon (n=1 in this case)S୧ : Stock price of the ith stock in the basketB୧ : Knock-out barrier of the ith stock in the basketProblem DescriptionJapan Equity-linked Power Reverse Dual Target Accrual Note (JRD): The JRD is a note targetedtowards Japanese investors wishing to gain exposure to US equity markets. Investors can alsogain exposure to FX via the “power reverse” feature. The note pays off a coupon based onthe best performing stock in a basket of 5 stocks. A detailed definition of terms and productspecifications can be found in the following sections.Definition of terms:Power Reverse Dual: Reverse dual notes offer coupons in domestic currency based on aforeign interest rate. Power reverse dual notes offer FX exposure and are suitable when youexpect the domestic/foreign rate to decline.The JRD note offers a twist on this standard product by allowing investors to gain equityexposure instead of just interest rates (which are currently at historic lows and not attractive).Moreover, the JPY is currently very strong. However given the state of the Japanese economyrelative to the US economy, we expect the JPY to decline - thereby the “power” feature inthe note, which exposes the investor to the FX rate, enhances the yield if the yen declines.Equity Linked Note: This note pays a coupon on a principle of ¥100 based on the bestperforming stock from a basket of 5 stocks.Up and Out Barrier: If any of the stocks in the underlying basket breach their preset barriers,the note terminates immediately, and investors a paid a bonus coupon to compensate themfor the early termination.
- 3. Product Specification: X JPY Issuer Investor Equity exposure Basket of Y*JPYUSD Y=Best of 5 5 stocksThe diagram above illustrates the basic setup of the JRD note. The mechanics are as follows: • Investor pays ¥X to the issuing bank • Principal is protected, but subject to FX risk. • On an annual basis, the investor receives ¥ Y*JPYUSD from the bank. Here Y is calculated as: ܵ − ܭ ௧ ݉ܽ ݔ൭ , 0൱ ∗ 100 ∀݅ ܵ బ ௧ Here ܵ is the stock prices of stock i on each observation day; ܵ బ is the stock ௧ ௧ price of stock i at the beginning of the coupon interval and ܭ is the strike price for stock i The JPYUSD exchange rate is calculated on the day of any cashflow. If any of the stocks ܵ in the basket hits their “knock out barrier” ܤ , the note is • ௧ • terminated, and the investor is paid a bonus coupon • ܤ is checked on each observation day. • After 1 year, if the note has not been redeemed or called, the note terminates and the investor received his final coupon on a notional of ¥100 (the basic pricing unit for the JRD) • On any of the observation dates, the investor can call the product. If the product is called the investor will be paid a coupon linked to the highest performing asset as of the exercise date (only in the Bermudan version of the JRD)Pricing a product with early exercise features is a complicated exercise when using MonteCarlo simulation. A simple way to do this would be to simulate entire trajectories, and thenat each point consider exercising versus holding on. However the price obtained by followingthis procedure will be biased upwards, as it is a “perfect foresight” estimate of the price. Inreality we do not know what the optimal exercise decision will be. Longstaff and Schwartz[1]present a technique based on a least squares regression that allows you to make a decisionfor early exercise based on the conditional expectation of the underlying asset price. Thistechnique requires the use of basis functions that are capable of expressing information
- 4. about the underlying prices that can then be used in an ordinary least squares regression togenerate conditional expectations. There is no hard and fast rule for the choice of basisfunctions, and we need to experiment with various kinds of functions to determine a good fit.In this paper we experiment with various kinds of functions (Monomials, Laguerre, Bessel,Hermite and Chebyshev). We pick a function that best captures the behavior of the JRDunder various scenarios.To focus on the simulation and the analysis of our product, we decide to use a simplegeometric Brownian motion model for the stock and FX dynamics, namely dS୧ = ሺr − y୧ ሻS୧ dt + σ୧ S୧ dW୧ ሺ1ሻ dQ = ሺrୢ − r ሻQdt + σ୕ QdW୕ ሺ2ሻwith the correlation matrix for the Brownian motions (5 stocks, and 1 FX rate which isassumed to be uncorrelated to any of the stocks hence has zero entries at all places except a 1 ρଶଵ ρଷଵ ρସଵ ρହଵ 0‘correlation’ of 1 with itself) ρ 1 ρଷଶ ρ ρହଶ 0 ۇρଵଶ ρଶଷ 1 ρସଶ ρହଷ 0ۊ ρ = ۈଶଷ ρ ധ ସଷ ρହସ 0ۋ ሺ3ሻ ρଷସ ଶସ ρଷସ 1 ρସହ ρଶହ ρସହ ρସହ 1 0 0 ۉ 0 0 0 0 1یhere r is the foreign risk-free rate, y୧ is the dividend yield for the ith stock, σ୧ is thevolatility of the ith stock, rୢ is the domestic risk-free rate and σ୕ is the volatility of theౚౣ౩౪ౙ ౙ౫౨౨ౙ౯ ౨ ౙ౫౨౨ౙ౯ FX rate. Considering the complexity in the structure, we will assume aconstant flat yield curve throughout.In this paper we consider we have a basket of 5 stocks whose correlation matrix we derivedfrom historical data. The correlation matrix along with the other parameters can be found inAppendix A.MethodologyHere we briefly describe how we priced the product including the least squares regression.Equity Linked option: Using cholesky decomposition, we generate correlated weinerprocesses. These processes are used to simulate stock trajectories. At each observation datewe consider a payout based on the best performing stock if the exercise cashflow is greaterthan the conditional expectation of the stock prices at that point (from the Least Squaresregression)Barrier Option: through matrices “stillAlive” and “breachingPoint” we keep of track of whenthe barrier was breached by any stock
- 5. Power Reverse Dual (PRDC): Each cashflow is multiplied by the corresponding FX rate whichis also simulated at each pointPricing on intermediate dates: This feature allows us to dynamically hedge the product. Tohandle this, we divide the maturity into intervals and calculate the number of intervalsremaining in each coupon.Least Squares Regression: At each observation date we determine the conditionalexpectation of future prices by regressing the current stock prices against some basisfunctions. This is further discussed in the “Basis Functions” section.BenchmarkingWe benchmarked our pricing function against several sources. 1) The basket call from Tavella [2] page 200: This example has a benchmark that was obtained by running a high quality finite differencing scheme on a basket of 3 assets. There are no barriers or other exotic features. The results obtained are shown in Table 1 Exercise Benchmark JRD Error Diff against Opportunities Pricer Benchmark 10 17.844 17.782 0.18 -0.062 15 17.969 17.891 0.13 -0.078 30 18.082 17.949 0.18 -0.133 Table 1: Benchmarking against basket of 3 assets from FD scheme[2] 2) Benchmarking against options with analytical results: To validate the JRD’s rainbow, barrier and FX features, we benchmark our product with simpler derivatives where analytical solutions are available. We implemented general closed form pricing models for options contingent to the maximum or the minimum of any number of assets using the result by Johnson[4]. With respect to the barrier feature, since our product has a discrete barrier which does not have a closed form solution, we benchmarked our result to a continuous barrier option’s analytical solution, with displaced barrier as suggested by Broadie and Glasserman[5].
- 6. Theoretical value JRD PricerRainbow option 13.243 13.288Barrier option (pricing from adjusted 3.4371 3.4512continuous barrier)Quanto option 4.1075 4.1409 Table 2: Benchmarking against Analytical option prices 3) Benchmarking against European structures: Another way to benchmark a pricing function is to consider the European version of the product and the Bermudan premium. This premium is discussed more extensively in the “European versus Bermudan structures” section.Results and DiscussionFor pricing the Bermudan JRD note, we utilized Monte Carlo simulations with the LeastSquares approach for early exercise. We also experimented with several variance reductiontechniques to accelerate the pricing over the naïve MC approach. We discuss thesetechniques briefly and then present our results.Quasi-Random Sequences: We implemented the Sobol sequence. In the naïve approach wesample randomly from a normal distribution. However this approach does not cover thesample space uniformly and as a result we experience a clustering effect. QRS numbersrectify this problem. In lower dimensions, QRS numbers cover the sample space relativelywell and therefore lead to more accurate pricing. A problem with QRS numbers is that unlikethe naive approach which has a clearly defined convergence law (and therefore a welldefined termination criterion), convergence is not as well defined. This is primarily becausethe theoretical error can be drastically different from the numerical error obtained. It istherefore good practice to calibrate to the naïve MC. The reader is referred to [2] for furtherdetails on this.Quasi-Random Sequences with Brownian Bridges: numbers generated from a QRS tend towork well in low dimensions. As the dimensionality of the problem increases however, thenumbers generated from quasi-random sequences tend to leave gaps. This reduces theacceleration obtained in higher dimensions. This can be exploited by using Brownian bridges.The numbers from the lower dimensions can be used at the more important points (end ofthe trajectory, mid point of trajectory etc). Each time we add a point using the Brownianbridge technique, the variance of the new point is reduced by half. We can therefore obtainsignificant acceleration in this way. To control the random numbers used in the constructionof the bridge we developed our own implementation of the bridge (rather than use the
- 7. default one implemented in MATLAB).Table 3 shows the results from under different scenarios. The parameters used to obtainthese results are shown in Appendix A. For the Least Squares regression we used 10 Hermitepolynomials on 3 cross products (ܵଵ , ܵଶ & ܵଵ ܵଶ) on the two best performing stocks in the ଶ ଶbasket at each observation date. Naïve Sobol Sobol with BBCycles Price Standard CPU Price Standard CPU Price Standard CPU Error Error Error 8 Exercise Opportunities10000 23.34 0.10656 74 23.31 0.06902 49 23.37 0.045109 4750000 23.34 0.06056 222 23.37 0.031173 265 23.41 0.020798 268100000 23.40 0.03715 196 23.38 0.020319 237 23.39 0.014715 252 16 Exercise Opportunities10000 23.81 0.09871 75 23.97 0.078865 96 23.95 0.056059 10250000 23.93 0.041052 544 24.01 0.028306 596 23.96 0.021163 615100000 24.00 0.03370 647 24.00 0.021172 1053 23.94 0.014697 864 32 Exercise Opportunities10000 24.40 0.09754 96 24.45 0.077832 113 24.38 0.056099 11050000 24.31 0.06085 807 24.40 0.03699 954 24.38 0.023484 1047100000 24.37 0.03605 984 24.37 0.022355 990 24.35 0.016417 1000 Table 3: Prices and errors obtained via naïve MC, MC with Sobol and MC with Sobol & Brownian bridgesDepending on the number of exercise opportunities, the JRD has dimensions ranging from 48to 192 (5 stocks + 1 fx rate times the # of exercise opportunities) in Table 3. The results showthat even though Sobol sequences do reduce the standard error, they are not as effective inhigher dimensions. Adding Brownian bridges however reduces the variance by more thanhalf in most of the cases.Latin Hypercube Sampling (LHS): extends the concept from stratified sampling (whichsamples from each ‘strata’ in the unit hypercube). LHS is similar except that it samples fromeach ‘row’ and ‘column’ in the multi dimensional hypercube such that no cell is repeated. Byincreasing the number of cells in the LHS, we can therefore obtain a sequence of randomnumbers that cover the sample space better than the standard sampling.Latin Hypercube Sampling (LHS) with Brownian Bridges: similar to the case with sobolsequences, we can construct Brownian bridges using random numbers generated from theLHS.
- 8. Using the same setup as before, we price the JRD using LHS. The results are presented intable 4. LHS LHS with BB Cycles Price Standard Error CPU Price Standard Error CPU 8 Exercise Opportunities 10000 23.56 0.077112 694 23.47 0.072128 484 50000 23.39 0.031476 2712 23.43 0.029398 2788 16 Exercise Opportunities 10000 23.93 0.081044 2026 23.91 0.059213 2137 50000 23.99 0.028629 10514 23.98 0.023684 9589 32 Exercise Opportunities 10000 24.39 0.083112 7163 24.40 0.045832 6560 50000 24.39 0.028626 38807 24.39 0.028538 37693 Table 4: Prices and errors obtained via MC with LHS and MC with LHS & Brownian bridgesTable 4 shows that while LHS does offer some acceleration, it does not work as well as Sobolsequences even when we add Brownian bridges. One explanation for this is sampling in theLHS scheme is done by drawing one sample from each cell. Therefore this approach isexpected to work best when the correlation between dimensions is minimal. Apart from theFX rate, the component stocks in the JRD are correlated with each other. Also the payofffunction depends on the entire trajectory and this introduces a complex correlation acrossthe observation dates. This correlation reduces the efficiency of LHS as a variance reductiontechnique for the JRD.One last point to note is that the CPU time for the LHS is very high. This can be explained bythe algorithm used to generate these samples. We used the MATLAB implementation of theLHS. As is evident, the LHS’ computational requirements are far higher than any of the othertechniques. Because of the computational cost we did not run a sample with 100,000cycles/samples as we did in the other schemes.From the JRD price, we generated an “exercise boundary” by calculating the maximum of theunderlying stock prices when the option was exercised. This boundary is shown in Fig 1.
- 9. Fig 1: Exercise Boundary versus the observation timeEuropean versus Bermudan structuresAn interesting study of the JRD’s behavior arises when we calculate the Bermudan premium.In calculating the premium there are a few factors to consider: • Dividend Yield: In the presence of dividends, options with early exercise features can become valuable. We therefore try a number of different dividend yields to determine how the pricing of the JRD behaves with respect to the dividend yield of the underlying basket of stocks. We consider constant and continuous dividend yields that are the same for all stocks for simplicity. The Bermudan premium as a fraction of the European price is shown in Figure 2 for a range of dividend yields (0 to 0.2) with 32 observation dates/exercise opportunities.
- 10. 0.22 Bermudan Premium as fraction of European Price 0.2 Laguerre Monomial 0.18 Hermite 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 Dividend Yield Fig 2: Bermudan premium versus Dividend Yields for the JRD Figure 2 shows the premiums obtained by using Laguerre (7 polynomials), Hermite (10 polynomials) and Monomials (7 monomials) as basis functions with 7 cross products (refer to Appendix B). We also experimented with Chebyshev polynomials of the 1st and 2nd kind, and Bessel equations of varying orders. However these functions performed poorly in the least squares regression, returning negative premiums at lower dividend yields. The interested reader is referred to Appendix C for plots showing the premiums obtained from these polynomials.• Barrier Level & Rebate: Analytically, there is a complex interaction between the barrier level and rebates and the Bermudan premium. When the barrier level is set above the rebate, then we expect a significant premium, as it is valuable to exercise when any of the underlying stocks approaches the barrier. When the reverse is true, the behavior depends on the level of the barrier. If the barrier is set very low, then the chances that both the European and Bermudan versions get knocked out early are higher, and we do not expect a big premium. However when the barrier is set high, then the premium should become more valuable. For a barrier level B୧ of 150% of the spot price for each stock, we evaluate the Bermudan premium of the JRD with 32 observation dates/exercise opportunities. Figure 3 shows the premium versus the rebate.
- 11. 0.2 Laguerre Bermudan Premium as fraction of European Price Monomial Hermite 0.15 0.1 0.05 0 -0.05 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 Rebate Fig 3: Bermudan premium versus Rebate for the JRD Figure 3 demonstrates an important point. When the rebate level is set higher than the barrier (i.e. it is advantageous to hit the barrier), the premium is low (negative in Figure 3). We know that the Bermudan premium cannot be below the European value, as we can always choose note to early exercise. We conjecture that the negative premiums observed are due to the regression function that we implement. In our implementation we only consider the top two stocks at each observation dates in order to reduce computational costs. If we early exercise based on these stocks, it is possible that the European version of the JRD becomes more valuable due to the barrier being breached by some other stock in the basket. To correct for this, we experimented with different cross products and also tried including all stocks prices in the regression. We were unable to determine a regression function that worked well for this particular case. It was also observed that including all stocks actually deteriorates the regression, and the interested reader is referred to Appendix C for a figure that demonstrates the premium obtained in this case. An added complexity is the FX dynamics which can also influence the exercise decision. For our case, we can simply set the price equal to the price of a European version when the rebate is greater than the barrier (we do not expect the premium to be very high in this case anyway). The uncertainty introduced by the barrier here should be dealt with through more sophisticated schemes such as Finite Differences.Basis FunctionsThe quality of the Least Squares Monte Carlo can be greatly impacted by the choice of basisfunctions. Simple polynomials may not be adequate for use in the least squares regression.
- 12. We therefore experimented with a number of polynomials including Laguerre, Bessel(varying orders), Chebyshev (1st and 2nd kinds) and Hermite functions. Figure 4 shows therebate generated through different basis functions. 0.2 Bermudan Premium as fraction of European Price 0.15 0.1 0.05 Laguerre Monomial Hermite 0 Chebyshev1st Chebyshev2nd Bessel -0.05 1 2 3 4 5 6 7 Number of Basis Functions Fig 4: Bermudan premium versus number of basis functions for the JRDFor the JRD, we have a basket of 5 correlated stocks. So our choice of basis functions shouldalso allow for the correlation between these stocks. For computational economy we onlyconsidered running regressions on the two best performing stocks and taking various crossproducts of these stocks. A definition of the cross products can be found in Appendix B. Wefound that the Chebyshev polynomials and the Bessel functions did not work well with a lownumber of cross products (negative premiums generated). Figure 5 demonstrates the impactof changing the number of cross products on Laguerre and Hermite polynomials.Note that the “monomial” basis function was taken from [2], and we do not change thenumber of cross products in this, therefore it is simply one value.
- 13. 0.18 Bermudan Premium as fraction of European Price 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 Laguerre Monomial 0 Hermite -0.02 1 2 3 4 5 6 7 Number of cross products Fig 5: Bermudan premium versus number of cross products for the JRDAnother check on the use of different basis functions is to consider the impact of changingthe strike. We do not expect big differences in premiums depending on whether the JRD is inthe money or out of the money. Basis functions that describe this behavior are more suitedfor pricing the JRD dynamically. Figure 6 shows premium for Laguerre, Monomials andHermite functions (the others do not produce valid results). 0.18 Bermudan Premium as fraction of European Price 0.16 0.14 0.12 0.1 0.08 Laguerre Monomial 0.06 Hermite 0.04 0.94 0.96 0.98 1 1.02 1.04 1.06 Normalized Strike Fig 6: Bermudan premium versus normalized strike for the JRDFigure 6 demonstrates that Monomials and Hermite functions produce the most consistentresults overall. In the pricing numbers presented in tables 3 and 4, we use 10 Hermitepolynomials.
- 14. HedgingIn this section we present two techniques for hedging the JRD 1) Dynamic Hedging and 2)Static Hedging. In dynamic hedging we simulate various trajectories for the underlying stocks,and at each point compute the delta. Based on the delta we can form a replicating portfoliothat can hedge our short position in the product (assuming we sold the product). By contraststatic hedging assumes that we do not change our hedging portfolio over the lifetime of theproduct. We therefore need to mimic the JRD’s early exercise and FX features, and weachieve this by using a combination of barrier and quanto options.Dynamic Delta HedgingWe follow a dynamic delta hedging strategy and compute the cost of hedging based on thechange in delta at each time step. This procedure is outlined in greater detail in Hull [3].To test this strategy we first simulate a set of asset trajectories (without loss of generality weassume the FX rate is fixed). We will test the delta hedging scheme as if these simulatedtrajectories are out of sample realized prices. We choose to carry out the hedging usingMonte Carlo, and in particular using the CV approach. The other two alternatives (i.e.,Likelihood Ratio and Pathwise Derivative) require us to compute partial derivatives, which isoutside the scope of our investigation.Figure 7 shows some sample trajectories. Note that the prices here are the actual stockprices from Appendix A (i.e not the normalized stock prices). Sample Realized Trajectories 350 300 250 200 150 100 50 0 0 2 4 6 8 10 12 14 16 18 Fig 7: Bermudan premium versus number of cross products for the JRDStarting from time zero and moving forward in time, we calculate the unperturbed derivativeprice P୳ and five perturbed prices P୲ଵ,ଶ,ଷ,ସ,ହ. The unperturbed price is calculated simply byusing the MC pricing function with maturity and spot prices updated for each time step. For
- 15. each of the perturbed prices, we change the corresponding spot price by multiplying it byሺ1 + ϵሻ where ϵ is small. Note that in calculating the 6 prices (perturbed or unperturbed),we have used the same set of Wiener processes for the Monte Carlo simulations. Thisensures that the error converges. The individual deltas are then approximated by taking theratio of P୲୧ − P୳ Δ୧ = , ∀i ϵP୳We rebalance the replicating portfolio one period later by changing our unit of stock holdingsto these one-period-lagged deltas. The rebalancing is self-financed by changing the positionin a USD, and in turn a JPY, cash account (recall that in this case the FX rate is fixed). Byrepeating this procedure until knock-out or expiry, we obtained a cost (or profit) of hedgingin JPY. If the hedging has been effective, we would expect that the discounted cost shouldequal the derivative price.Figure 8 shows a histogram with the distribution of hedging costs for multiple simulations.The histogram is close to a normal distribution. However, we notice that the average cost ofhedging is as high as JPY2700. Histogram of Cost of Delta Hedging 25 20 15 10 5 0 -8 -6 -4 -2 0 2 4 Cost in JPY 4 x 10 Fig 8: Cost of hedging (JPY)To investigate why delta hedging does not work as expected, we first repeat the experimentusing one tenth of the previous volatility levels. Figure 9 shows a histogram of the cost in thiscase.
- 16. Histogram of Cost of Delta Hedging 25 20 15 10 5 0 -1.5 -1 -0.5 0 0.5 1 1.5 Cost in JPY 4 x 10 Fig 9: Cost of hedging (One Tenth volatility)The average cost of hedging drops from JPY2700 to JPY800. Furthermore, if we look at thechange in the deltas in different time steps, as shown in the following table, the value ofdelta can change abruptly at different points in time.Time Step ઢ ઢ ઢ ઢ ઢ1 0.0326 0.1617 0.7020 0.1110 3.78212 0.0348 0.1829 5.7690 3.7583 1.09303 0.0410 0.1817 0.4511 0.1464 -16.90504 0.0378 0.2077 1.4777 0.1937 1.4602 Table 5: Delta with respect to individual stocksTaking the two pieces of evidence into account, we make the following postulation regardingthe poor performance of delta hedging for JRD. The payoff of JRD depends on the return ofthe best performer in the basket. By this nature, the value of the JRD is a highly non-linearfunction of the underlying asset prices. As a didactic example, suppose that by the CVapproach we perturb the price of Asset 1 and found that Δଵ = 0.03 and suppose that atthis price level Asset 1 is the best performer. When the price of Asset 1 changes further,however, it ceases to be the best performer and the JRD value is determined by the return ofAsset 4. As a consequence Δଵ is not a smooth function of S1. This agrees with the abruptlychanging deltas we saw in the table. In other words, the gamma of JRD is so high that deltahedging simply cannot capture the instantaneous risk. This also agrees with the fact thatunder lower volatility the hedging performance is marginally improved, since under lowervolatility a best performer has a higher probability to remain the best performer, and viceversa.
- 17. This analysis suggests that hedging may better be achieved through a static approach, wherewe do not need to rebalance and incur hedging costs at each observation date. This isdiscussed in the next section.Static HedgingThe exotic features of this product include: Rainbow option (best performance of the five underlying) Exchange option (paid in JPY) Knock out Barrier option (can be triggered by any of the five assets) First touch option (rebate)Due to the nature of complexity of the product, it is quite difficult to achieve perfect hedgingby implementing a static hedging strategy. Therefore, we employ an approximate statichedging using Mean Monte Carlo (MMC). This procedure is equivalent to partially hedgingthe claim using a portfolio of simpler assets. If the residual risk is deemed appropriate or canbe reduced to a suitable level, the static portfolio can be employed in place of complexdynamic hedging procedures (for further details refer to Pellizzari 2001 [6]).Here we denote the final payoff of the coupon at ܶ = ݐas ݂ሺܵଵ் , ܵଶ் … ்ܵ ሻ . We then { ்ܯሺ݆ሻ}ୀଵ,ଶ… , ݓℎ݁ ்ܯ ݁ݎሺ݆ሻ = ݂ሺܧሺܵଵ் ሻ, ܧሺܵଶ் ሻ, … ்ܵ , … ܧሺ்ܵ ሻሻchoose a combination of low dimensional instrumentsSuch that the variance of ݂ሺܵଵ் , ܵଶ் … ்ܵ ሻ − ߚ ்ܯሺ݆ሻ ୀଵis minimized. However, the barrier in our product can be triggered by any asset. So it is quitedifferent from a combination of multiple barrier options each of which has a singleunderlying asset with its own barriers.As the base price is set at time 0, the strike prices for all the options are at the money. Thehedging portfolio then consists of: Six month at the money European Barrier options on three of the most volatile underlying stocks Six month at the money European Exchange Option between USD and JPY Six month European first touch options on three of the most volatile underlying stocksThe trading strategy is to hold the hedging portfolio at the beginning, with the weightdetermined by the following procedure, and remaining money is put into a risk free account.For each trajectory, if the product hit the barrier, we clear all of our positions, pay the rebate,
- 18. and put the remaining, if any left, into risk free account till the end. If the product is earlyexercised, we clear all of our positions as well and pay the option payoff, and put theremaining, if any left, into risk free account till the end. We took 10,000 trajectories andrecorded the payoff profit and loss for each one. We then chose our weight of holding oneach component such that the total variance of final profit and loss is minimized. Finally weused the weights calculated above to test the hedging on another 10,000 trajectories. Theresult is summarized in Table 6. % % % MMC Hedging -0.22627 -0. 13964 -0.014769 0.077624Without Hedging -0.30433 -0.23425 -0.14161 0.15011 Table 6: Simulation results with MMC hedging Fig 10: PnL with MMC hedging
- 19. Fig 11: PnL without MMC hedging (note the big positive premium and fat tail)We can find that although the hedging portfolio is not completely hedged, the hedgingportfolio weighted by the Mean Monte Carlo technique using 10,000 simulations is betterdistributed than the unhedged portfolio results as shown in Figures 10 and 11 (both VAR andvolatility are much smaller).As the market marker, it would not be hard to get these instruments from the OTC market.If we cannot find a barrier option with the desired barrier in the market, we can insteadreplicate this single barrier option with a combination of vanilla index options with differentstrikes, using the hedging techniques suggested by Carr and Chou [7].It is clear that the static hedging appears to work better for the JRD, due to the hedging costsinvolved with dynamic hedging. The dimensionality of the problem is the critical factor here.With a high dimension product, we are faced with very high delta’s and this is difficult tohedge with only first order hedging. If we have access to a market with other exotic optionson single assets, we can however create a hedging portfolio that minimizes our variance asdemonstrated by the MMC hedging technique.Final RemarksIn this paper we have demonstrated pricing and hedging of a complicated structured productusing the Least Squares Monte Carlo approach. We saw that as the dimensionality of theproduct increases, we are forced to think of innovative techniques to achieve the desiredaccuracy from the pricing function. In this paper we achieved significant variance reductionby using a combination of Sobol numbers and Brownian Bridges. To get comfortable with the
- 20. price of the product, we also benchmarked our pricing to several analytical results and a highquality numerical result. We also demonstrated two different hedging techniques for thisproduct and showed that it is cheaper to use static hedging for the JRD note.The MATLAB implementation of this product can be found in Appendix D for the interestedreader.References[1] Longstaff and Schwartz, “Valuing American options by simulation: a simple least-squares approach”[2] Tavella, Domingo A., Quantitative Methods in Derivatives Pricing[3] Hull, John C., Options, Futures and Other Derivatives (6th Edition)[4] Johnson, Herb “Options on the Maximum or the Minimum of Several Assets.” The Journal of Financial and Quantitative Analysis, Vol. 22, 1987[5] Broadie and Glasserman[6] Pellizzari, P. "Static Hedging of Multivariate Derivatives by Simulation." European Journal of Operational Research (2005): 507-19. Web.[7] Carr, Peter and Chou, Andrew “Breaking Barriers.” Risk September 1987[8] Derman, E., D. Ergener, and I. Kani. Static Options Replication. Goldman, Sachs & Co.[9] Jäckel, Peter. Monte Carlo Methods in Finance. Chichester, West Sussex, England: J. Wiley, 2002. Print.[10] Piterbarg, V. “A Multi-currency Model with FX Volatility Skew.” (February 7, 2005). Available at SSRN: http://ssrn.com/abstract=685084[11] Pricing Power Reverse Dual Currency Notes. Available at http://www.fincad.com/derivatives-resources/articles/price-power-reverse-dual-curr ency.aspx
- 21. Appendix A – Parameters UsedThe parameters used to price our final product:Stocks used: Apple, Citi, P&G, American Express and MicrosoftCorrelation Matrix:1.0000 0.2957 0.3729 0.1930 0.2438 00.2957 1.0000 0.3819 0.3738 0.5604 00.3729 0.3819 1.0000 0.2846 0.3097 00.1930 0.3738 0.2846 1.0000 0.2846 00.2438 0.5604 0.3097 0.2846 1.0000 00 0 0 0 0 1(note the final row and column refers to the FX rate, which is assumed to be uncorrelatedwith the underlying stocks)Volatilities: [0.15;0.2;0.23;0.3;0.1];Spot Prices: [250;40;23;60;3.8];Dividend Yields: [0.05;0.02;0.09;0.01;0.02];Barriers: spots*1.2 (20% increase in any stock)Rebate: 30%Maturity: 1 yearCoupon maturity : 1 year (annual)Foreign Interest Rates: 5%Domestic Interest Rates: 1%FX volatility: 15%Initial FX Rate: 100
- 22. Appendix B – Basis Functions and Cross ProductsLaguerre Polynomials1 12 −1 + ݔ 1 ଶ ሺ2 + ݔ4 − ݔሻ 23 1 ሺ− ݔଷ + 9 ݔଶ − 186 + ݔሻ 64 1 ସ ሺ ݔ61 − ݔଷ + 72 ݔଶ − 9642 + ݔሻ 245 1 ሺ− ݔହ + 25 ݔସ − 200 ݔଷ + 600 ݔଶ − 600021 + ݔሻ 1206 1 ሺ ݔ63 − ݔହ + 450 ݔସ − 2400 ݔଷ + 5400 ݔଶ − 4320027 + ݔሻ 7207Hermite (probabilists) Polynomials1 12 ݔ3 ݔଶ − 14 ݔଷ − 3ݔ5 ݔସ − 6 ݔଶ + 36 ݔହ − 10 ݔଷ + 15ݔ7 ݔ51 − ݔସ + 45 ݔଶ − 158 ݔ12 − ݔହ − 105 ݔଷ + 105ݔ9 ݔ012 + ݔ82 − ଼ ݔସ − 420 ݔଶ + 10510 ݔଽ − 36 ݔ873 + ݔହ − 1260 ݔଷ + 945ݔChebyshev (1st kind) Polynomials1 12 ݔ3 2 ݔଶ − 14 4 ݔଷ − 3ݔ5 8 ݔସ − 8 ݔଶ + 16 16 ݔହ − 20 ݔଷ + 5ݔ7 32 ݔ84 − ݔସ + 18 ݔଶ − 18 64 ݔ211 − ݔହ + 56 ݔଷ − 7ݔ
- 23. 9 128 ݔ061 + ݔ652 − ଼ ݔସ − 32 ݔଶ + 110 256 ݔଽ − 576 ݔ234 + ݔହ − 120 ݔଷ + 9ݔChebyshev (2nd kind) Polynomials1 12 2ݔ3 4 ݔଶ − 14 8 ݔଷ − 4ݔ5 16 ݔସ − 12 ݔଶ + 16 32 ݔହ − 32 ݔଷ + 6ݔ7 64 ݔ08 − ݔସ + 24 ݔଶ − 18 128 ݔ291 − ݔହ + 80 ݔଷ − 8ݔ9 256 ݔ042 + ݔ844 − ଼ ݔସ − 40 ݔଶ + 110 512 ݔଽ − 1024 ݔ276 + ݔହ − 160 ݔଷ + 10ݔBessel Functions: We use the MATLAB implementation of the Bessel functions of variousordersMonomials (taken from Tavella[2])1 ܵଵ + ܵଵ + ܵଵ + ܵଵ + ܵଵ ଶ ଷ ସ ହ2 ܵଶ + ܵଶ + ܵଶ + ܵଶ + ܵଶ ଶ ଷ ସ ହ3 ܵଵ ܵଶ + ܵଵ ܵଶ + ܵଵ ܵଶ + ܵଵ ܵଶ ଶ ଷ ସ4 ܵଵ ܵଶ + ܵଵ ܵଶ + ܵଵ ܵଶ ଶ ଶ ଶ ଶ ଷ5 ܵଵ ܵଶ + ܵଵ ܵଶ ଷ ଷ ଶ6 ܵଵ ܵଶ ସ7 1For the basis functions (Laguerre, Hermite and Laguerre), we use the following crossproducts:1 ܵଵ2 ܵଶ3 ܵଵ ܵଶ4 ܵଵ ܵଶ ଶ5 ܵଵ ܵଶ ଷ6 ܵଵ ܵଶ ଶ ଶ7 ܵଵ ܵଶ ଷ ଷWe also tried cross products with all 5 stocks included
- 24. 1 ܵଵ ଶ2 ܵଶ ଶ3 ܵଷ ଶ4 ܵସ ଶ5 ܵହ ଶ6 ܵଵ ܵଶ7 ܵଵ ܵଶ ଶ8 ܵଵ ܵଷ9 ܵଵ ܵସ10 ܵଵ ܵହ11 ܵଶ ܵଷ12 ܵଶ ܵସ
- 25. Appendix C – Misc Graphs 0.3 Bermudan Premium as fraction of European Price 0.2 0.1 0 -0.1 -0.2 -0.3 Chebyshev1st -0.4 Chebyshev2nd Bessel -0.5 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 Div Yield Fig C.1: Bermudan premium with varying div yieldsFrom Fig C.1 it is clear that the Chebyshev polynomials and Bessel functions do not work wellwith low dividend yields and generate negative premiums. 0.1 Bermudan Premium as fraction of European Price 0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 Chebyshev1st Chebyshev2nd -0.7 Bessel Hermite with all cross products -0.8 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 Rebate Fig C.2: Bermudan premium with varying rebatesFig C.2 shows that Chebyshev polynomials and Bessel functions do not work well withbarriers. In the paper we talked about how we obtained negative premiums even withHermite, laguerre and monomials when the rebate was set high above the barrier. Wehypothesized that this was due to the other components of the basket which were not
- 26. included in the least squares regression. However Fig C.2 shows that even when we includemore cross products which cover all the assets, we obtain negative premiums. In fact in thiscase the regression does not work well at all, with consistent negative premiums. Thissuggests that either the basis functions and cross products need to be more complex, or thatthe LSMC technique cannot handle the complexity introduced by the barriers and thedimensionality of the product, and we must use other schemes such as Finite Differencing. 0.1 Bermudan Premium as fraction of European Price 0.05 0 Chebyshev1st Chebyshev2nd -0.05 Bessel -0.1 -0.15 -0.2 -0.25 -0.3 -0.35 0.94 0.96 0.98 1 1.02 1.04 1.06 Normalized Strike Fig C.3: Bermudan premium with varying rebatesFig C.3 shows the Bermudan premium against normalized strikes when using Chebyshev andBessel functions. Bessel functions do produce a positive premium, but the premium is muchlower than we expect. We therefore disregard these functions.
- 27. Appendix D – MATLAB implementation%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Main pricing function (includes all variance reduction% techniques and also prices the European version)%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%function MCResults=jrd_LSMC(maturity, vols, divYields, foreignRate,...spots, strikes,observation_frequency, cycles, corMatrix,basketSize,extraAssets, ...highBarrier,rebate,domesticRate,fxVol,fxRate,batches,MCType,useBrownianBridge,... basisType,basisFns,crossProducts,productType)MCResults=zeros(3,1);% parametersinterval = 1/observation_frequency;payoffs = zeros(batches,1);cholMatrix=chol(corMatrix);strikes = strikes./spots; %scale strike for ease of use in basis functionstime_steps = observation_frequency*maturity + 1;dimensions=(basketSize + extraAssets + 1)*(time_steps-1);highBarrier= highBarrier./spots;if strcmp(MCType,sobol) sobol_obj=sobolset(dimensions,Skip,1e3);endfor i=1:batches if strcmp(MCType,sobol) sobol_seq=sobol_obj((i-1)*cycles+1:i*cycles,:); sobol_norm=norminv(sobol_seq); sobol_norm_seq=reshape(sobol_norm,[cycles,time_steps-1,basketSize + extraAssets + 1]); elseif strcmp(MCType,lhs)
- 28. lhs_unformatted=lhsdesign(cycles,dimensions,smooth,off,criterion,correlation); lhs_unformatted=norminv(lhs_unformatted); lhs=reshape(lhs_unformatted,cycles,time_steps-1,basketSize +extraAssets + 1); end % work with normalizes stock prices and fx rates, hence period 0 values % are 1 trajectories = ones(cycles, time_steps, basketSize + extraAssets); fxTrajectories = ones(cycles,time_steps); trajectoriesW = ones(cycles, time_steps, basketSize + extraAssets); fxTrajectoriesW = ones(cycles,time_steps); % generate weiner processes if useBrownianBridge % steps must be in 2^k + 1 format! currInterval = maturity; % generate beginning and end points randSetIndex = 1; beginW = zeros(cycles,basketSize + extraAssets); beginFxW = zeros(cycles,1); trajectoriesW(:,1,:) = reshape(beginW,cycles,1,basketSize +extraAssets); fxTrajectoriesW(:,1) = beginFxW; if strcmp(MCType,sobol) endW =(cholMatrix*sobol_norm(:,randSetIndex:randSetIndex+basketSize+extraAssets-1))*sqrt(currInterval); randSetIndex = randSetIndex + basketSize+extraAssets; endFxW =sobol_norm(:,randSetIndex:randSetIndex)*sqrt(currInterval); randSetIndex = randSetIndex + 1; elseif strcmp(MCType,lhs) endW =(cholMatrix*lhs_unformatted(:,randSetIndex:randSetIndex+basketSize+extraAssets-1))*sqrt(currInterval); randSetIndex = randSetIndex + basketSize+extraAssets;
- 29. endFxW =lhs_unformatted(:,randSetIndex:randSetIndex)*sqrt(currInterval); randSetIndex = randSetIndex + 1; end trajectoriesW(:,time_steps,:) =reshape(endW,cycles,1,basketSize + extraAssets); fxTrajectoriesW(:,time_steps) = endFxW; bbInterval = interval; while (currInterval > interval) bbIndex = 1; trajIndex = 1; while (bbIndex*currInterval <= maturity) if strcmp(MCType,sobol) currZ =(cholMatrix*sobol_norm(:,randSetIndex:randSetIndex+basketSize+extraAssets-1)); randSetIndex = randSetIndex + basketSize+extraAssets; currFxZ = sobol_norm(:,randSetIndex:randSetIndex); randSetIndex = randSetIndex + 1; elseif strcmp(MCType,lhs) currZ =(cholMatrix*lhs_unformatted(:,randSetIndex:randSetIndex+basketSize+extraAssets-1)); randSetIndex = randSetIndex + basketSize+extraAssets; currFxZ =lhs_unformatted(:,randSetIndex:randSetIndex); randSetIndex = randSetIndex + 1; end variance = currInterval/4; currW = 0.5*(beginW+endW) + sqrt(variance)*currZ; currFxW = 0.5*(beginFxW+endFxW) + sqrt(variance)*currFxZ; trajIndex = trajIndex +((time_steps-1)/(maturity/(currInterval/2))); trajectoriesW(:,trajIndex,:) =reshape(currW,cycles,1,basketSize + extraAssets); fxTrajectoriesW(:,trajIndex) = currFxW; bbIndex = bbIndex + 1;
- 30. beginW = endW; beginFxW = endFxW; trajIndex = trajIndex +((time_steps-1)/(maturity/(currInterval/2))); endIndex =((time_steps-1)/(maturity/currInterval))*bbIndex + 1; if endIndex <= time_steps endW =reshape(trajectoriesW(:,endIndex,:),cycles,basketSize + extraAssets); endFxW = fxTrajectoriesW(:,endIndex); end end currInterval = currInterval/2; beginW = zeros(cycles,basketSize + extraAssets); beginFxW = zeros(cycles,1); trajIndex = ((time_steps-1)/(maturity/currInterval)) + 1; endW =reshape(trajectoriesW(:,trajIndex,:),cycles,basketSize + extraAssets); endFxW = fxTrajectoriesW(:,trajIndex); bbInterval = bbInterval/2; end % free memory clear beginW endW beginFxW endFxW currW currFxW; % generate trajectories for j=2:time_steps for k=1:basketSize + extraAssets trajectories(:,j,k) =trajectories(:,j-1,k).*exp((foreignRate... -divYields(k)-0.5*(vols(k)^2))*interval +vols(k)*(trajectoriesW(:,j,k)-trajectoriesW(:,j-1,k))); end fxTrajectories(:,j) =fxTrajectories(:,j-1).*exp((domesticRate... -foreignRate-0.5*(fxVol^2))*interval +fxVol*(fxTrajectoriesW(:,j)-fxTrajectoriesW(:,j-1))); end clear trajectoriesW fxTrajectoriesW; else for j=2:time_steps
- 31. if strcmp(MCType,sobol)sobol_eachtimeStep=reshape(sobol_norm_seq(:,(j-1),:),[cycles,basketSize + extraAssets + 1]); z=sobol_eachtimeStep(:,1:basketSize +extraAssets)*cholMatrix; fxZ=sobol_eachtimeStep(:,basketSize + extraAssets +1:basketSize + extraAssets + 1); elseif strcmp(MCType,lhs) lhs_eachtimeStep=reshape(lhs(:,(j-1),:),[cycles,basketSize + extraAssets + 1]); z=lhs_eachtimeStep(:,1:basketSize +extraAssets)*cholMatrix; fxZ=lhs_eachtimeStep(:,basketSize + extraAssets +1:basketSize + extraAssets + 1); else z=(cholMatrix*normrnd(0,1,basketSize +extraAssets,cycles)); fxZ=randn(cycles,1); end % generate trajectories for k=1:basketSize + extraAssets trajectories(:,j,k) =trajectories(:,j-1,k).*exp((foreignRate... -divYields(k)-0.5*(vols(k)^2))*interval +vols(k)*sqrt(interval)*z(:,k)); end fxTrajectories(:,j) =fxTrajectories(:,j-1).*exp((domesticRate... -foreignRate-0.5*(fxVol^2))*interval +fxVol*sqrt(interval)*fxZ); end end clear z sobol_norm_seq sobol_norm lhs_unformatted lhs; % free up memory % work backwards cfMatrix = zeros(cycles, time_steps-1); cfMatrixPlusRebate= zeros(cycles, observation_frequency*maturity); stillAlive=zeros(cycles, observation_frequency*maturity);
- 32. % determine final payoff based on initial composition of the basket currPrice = reshape(trajectories(:,time_steps,1:basketSize),cycles,basketSize); cfMatrix(:,time_steps-1) = (max(max(currPrice... -repmat(strikes(1:basketSize),cycles,1),0),[],2)).*fxTrajectories(:,time_steps-1); alive=ones(cycles,1); % determine the trajectories where the barriers are not breached for m=1:observation_frequency*maturity temp=reshape(trajectories(:,m+1,:),cycles,basketSize); alive=alive.* min((repmat(highBarrier,cycles,1)-temp)>0,[],2); stillAlive(:,m)=alive; end breachingPoint=[ones(cycles,1),stillAlive(:,1:(end-1))]-stillAlive;cfMatrix(:,time_steps-1)=cfMatrix(:,time_steps-1).*stillAlive(:,time_steps-1); % pay rebate at the point where barriers were breached cfMatrixPlusRebate = cfMatrix + rebate.*breachingPoint; % loop through to determine bermudan value if strcmp(productType,bermudan) for j=time_steps-1:-1:1 % determining current exercise value currPrice = reshape(trajectories(:,j+1,1:basketSize),cycles,basketSize); exerciseCF = (max(max(currPrice -repmat(strikes(1:basketSize),cycles,1),0),[],2)).*fxTrajectories(:,j+1); exerciseCF = exerciseCF.*stillAlive(:,j); % cannot exercise ifknocked out moneyness = exerciseCF>0; pvVector = zeros(cycles,1); for k=j+1:time_steps-1 % As all future cashflows have already been converted to % the % domestic currency discounting done in domestic rate
- 33. pvVector = pvVector +cfMatrixPlusRebate(:,k)*exp(-domesticRate*interval*(k-j)); end % only consider the two best performing assets for the least % squares regression sortedPrices =reshape(sort(trajectories(:,j+1,1:basketSize),3,descend),cycles,basketSize); currPrices = sortedPrices(:,1:2); if sum(moneyness)==0 condExpect = moneyness; else if strcmp(basisType,all) condExpect =LSR(sortedPrices,pvVector,basisType,basisFns,crossProducts,moneyness) else condExpect =LSR(currPrices,pvVector,basisType,basisFns,crossProducts,moneyness); end end cfMatrix(:,j) = moneyness.*((condExpect <exerciseCF).*exerciseCF); % If knocked out pay the rebate cfMatrixPlusRebate(:,j) = cfMatrix(:,j) +rebate*breachingPoint(:,j); % set CFs of j+1 to maturity zero if exercised at current time for k=j+1:time_steps-1 % If not exercised and not knocked out, keep old cashflows cfMatrix(:,k) = cfMatrix(:,k).*(cfMatrix(:,j)==0); cfMatrixPlusRebate(:,k) =cfMatrixPlusRebate(:,k).*(cfMatrix(:,j)==0); end end end % Discount the CFs from the CF matrix to get option value currPayoffs = zeros(cycles,1); for j=time_steps-1:-1:1
- 34. currPayoffs = currPayoffs +cfMatrixPlusRebate(:,j).*exp(-(domesticRate/observation_frequency)*j); end payoffs(i) = mean(currPayoffs)*fxRate; % Scale by current fxRateendclear trajectories fxTrajectories; % free memoryMCResults(1)=payoffs(1);MCResults(2) = std(payoffs);end%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Function to perform Least Squares Regression for LSMC%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%function condCF=LSR(currPrices, pvVector, fnType,numFns,crossProducts,moneyness)condCF=zeros(size(currPrices,1),1);crossTerms = zeros(size(currPrices,1),7);regressFns = numFns;if strcmp(fnType,simplest) basisFns = zeros(size(currPrices,1),numFns); for i=2:numFns basisFns(:,i) = currPrices.^(i-1); endelseif strcmp(fnType,sine) basisFns = zeros(size(currPrices,1),numFns); for i=2:numFns basisFns(:,i) = sin((i-1)*currPrices); endelseif strcmp(fnType,monomial) basisFns = zeros(size(currPrices,1),numFns); basisFns(:,1) = currPrices(:,1) + currPrices(:,1).^2 +currPrices(:,1).^3 ... + currPrices(:,1).^4 + currPrices(:,1).^5; if numFns >= 2 basisFns(:,2) = currPrices(:,2) + currPrices(:,2).^2 +currPrices(:,2).^3 ...
- 35. + currPrices(:,2).^4 + currPrices(:,2).^5; end if numFns >= 3 basisFns(:,3) = currPrices(:,1).*currPrices(:,2) +currPrices(:,1).*(currPrices(:,2).^2) ... + currPrices(:,1).*(currPrices(:,2).^3) +currPrices(:,1).*(currPrices(:,2).^4); end if numFns >= 4 basisFns(:,4) = (currPrices(:,1).^2).*currPrices(:,2) +(currPrices(:,1).^2).*(currPrices(:,2).^2) ... + (currPrices(:,1).^2).*(currPrices(:,2).^3); end if numFns >= 5 basisFns(:,5) = (currPrices(:,1).^3).*currPrices(:,2) +(currPrices(:,1).^3).*(currPrices(:,2).^2); end if numFns >= 6 basisFns(:,6) = (currPrices(:,1).^4).*currPrices(:,2); end if numFns >= 7 basisFns(:,7) = 1; endelseif strcmp(fnType,laguerre) crossTerms(:,1) = sum(laguerre(currPrices(:,1),numFns),2); crossTerms(:,2) = sum(laguerre(currPrices(:,2),numFns),2); crossTerms(:,3) =sum(laguerre(currPrices(:,1).*currPrices(:,2),numFns),2); crossTerms(:,4) =sum(laguerre((currPrices(:,1).^2).*currPrices(:,2),numFns),2); crossTerms(:,5) =sum(laguerre((currPrices(:,1).^3).*currPrices(:,2),numFns),2); crossTerms(:,6) =sum(laguerre((currPrices(:,1).^2).*(currPrices(:,2).^2),numFns),2); crossTerms(:,7) =sum(laguerre((currPrices(:,1).^3).*(currPrices(:,2).^3),numFns),2); basisFns = crossTerms(:,1:crossProducts); regressFns = crossProducts;
- 36. elseif strcmp(fnType,Chebyshev1st) crossTerms(:,1) = sum(Chebyshev1st(currPrices(:,1),numFns),2); crossTerms(:,2) = sum(Chebyshev1st(currPrices(:,2),numFns),2); crossTerms(:,3) =sum(Chebyshev1st(currPrices(:,1).*currPrices(:,2),numFns),2); crossTerms(:,4) =sum(Chebyshev1st((currPrices(:,1).^2).*currPrices(:,2),numFns),2); crossTerms(:,5) =sum(Chebyshev1st((currPrices(:,1).^3).*currPrices(:,2),numFns),2); crossTerms(:,6) =sum(Chebyshev1st((currPrices(:,1).^2).*(currPrices(:,2).^2),numFns),2); crossTerms(:,7) =sum(Chebyshev1st((currPrices(:,1).^3).*(currPrices(:,2).^3),numFns),2); basisFns = crossTerms(:,1:crossProducts); regressFns = crossProducts;elseif strcmp(fnType,Chebyshev2nd) crossTerms(:,1) = sum(Chebyshev2nd(currPrices(:,1),numFns),2); crossTerms(:,2) = sum(Chebyshev2nd(currPrices(:,2),numFns),2); crossTerms(:,3) =sum(Chebyshev2nd(currPrices(:,1).*currPrices(:,2),numFns),2); crossTerms(:,4) =sum(Chebyshev2nd((currPrices(:,1).^2).*currPrices(:,2),numFns),2); crossTerms(:,5) =sum(Chebyshev2nd((currPrices(:,1).^3).*currPrices(:,2),numFns),2); crossTerms(:,6) =sum(Chebyshev2nd((currPrices(:,1).^2).*(currPrices(:,2).^2),numFns),2); crossTerms(:,7) =sum(Chebyshev2nd((currPrices(:,1).^3).*(currPrices(:,2).^3),numFns),2); basisFns = crossTerms(:,1:crossProducts); regressFns = crossProducts;elseif strcmp(fnType,Hermite) crossTerms(:,1) = sum(Hermite(currPrices(:,1),numFns),2); crossTerms(:,2) = sum(Hermite(currPrices(:,2),numFns),2);
- 37. crossTerms(:,3) =sum(Hermite(currPrices(:,1).*currPrices(:,2),numFns),2); crossTerms(:,4) =sum(Hermite((currPrices(:,1).^2).*currPrices(:,2),numFns),2); crossTerms(:,5) =sum(Hermite((currPrices(:,1).^3).*currPrices(:,2),numFns),2); crossTerms(:,6) =sum(Hermite((currPrices(:,1).^2).*(currPrices(:,2).^2),numFns),2); crossTerms(:,7) =sum(Hermite((currPrices(:,1).^3).*(currPrices(:,2).^3),numFns),2); basisFns = crossTerms(:,1:crossProducts); regressFns = crossProducts;elseif strcmp(fnType,all) crossTerms(:,1) = sum(Hermite(currPrices(:,1).^2,numFns),2); crossTerms(:,2) = sum(Hermite(currPrices(:,2).^2,numFns),2); crossTerms(:,3) = sum(Hermite(currPrices(:,3).^2,numFns),2); crossTerms(:,4) = sum(Hermite(currPrices(:,4).^2,numFns),2); crossTerms(:,5) = sum(Hermite(currPrices(:,5).^2,numFns),2); crossTerms(:,6) =sum(Hermite(currPrices(:,1).*currPrices(:,2),numFns),2); crossTerms(:,7) =sum(Hermite((currPrices(:,1).^2).*currPrices(:,2),numFns),2); crossTerms(:,8) =sum(Hermite(currPrices(:,1).*currPrices(:,3),numFns),2); crossTerms(:,9) =sum(Hermite(currPrices(:,1).*currPrices(:,4),numFns),2); crossTerms(:,10) =sum(Hermite(currPrices(:,1).*currPrices(:,5),numFns),2); crossTerms(:,11) =sum(Hermite(currPrices(:,2).*currPrices(:,3),numFns),2); crossTerms(:,12) =sum(Hermite(currPrices(:,1).*currPrices(:,2),numFns),2)... +sum(Hermite(currPrices(:,2).*currPrices(:,3),numFns),2); basisFns = crossTerms(:,1:crossProducts); regressFns = 12;elseif strcmp(fnType,bessel) crossTerms(:,1) = sum(bessel(numFns,currPrices(:,1)),2); crossTerms(:,2) = sum(bessel(numFns,currPrices(:,2)),2);
- 38. crossTerms(:,3) =sum(bessel(numFns,currPrices(:,1).*currPrices(:,2)),2); crossTerms(:,4) =sum(bessel(numFns,(currPrices(:,1).^2).*currPrices(:,2)),2); crossTerms(:,5) =sum(bessel(numFns,(currPrices(:,1).^3).*currPrices(:,2)),2); crossTerms(:,6) =sum(bessel(numFns,(currPrices(:,1).^2).*(currPrices(:,2).^2)),2); crossTerms(:,7) =sum(bessel(numFns,(currPrices(:,1).^3).*(currPrices(:,2).^3)),2); basisFns = crossTerms(:,1:crossProducts); regressFns = crossProducts;end% basisFns(:,1) = basisFns(:,1).*moneyness;formattedBasisFns = zeros(sum(moneyness),regressFns);formattedPvs = zeros(sum(moneyness),1);counter = 1;for i=1:size(currPrices,1) if moneyness(i)==1 formattedBasisFns(counter,:) = basisFns(i,:); formattedPvs(counter,:) = pvVector(i,:); counter = counter + 1; endendb = regress(formattedPvs,formattedBasisFns);if strcmp(fnType,simplest) for i=1:numFns condCF = condCF + b(i)*currPrices.^(i-1); endelseif strcmp(fnType,sine) condCF = b(1); for i=2:numFns condCF = condCF + b(i)*sin((i-1)*currPrices); endelseif strcmp(fnType,monomial) condCF = condCF + b(1)*(currPrices(:,1) + currPrices(:,1).^2 +currPrices(:,1).^3 ... + currPrices(:,1).^4 + currPrices(:,1).^5);
- 39. if numFns >= 2 condCF = condCF + b(2)*(currPrices(:,2) + currPrices(:,2).^2 +currPrices(:,2).^3 ... + currPrices(:,2).^4 + currPrices(:,2).^5); end if numFns >= 3 condCF = condCF + b(3)*(currPrices(:,1).*currPrices(:,2) +currPrices(:,1).*(currPrices(:,2).^2) ... + currPrices(:,1).*(currPrices(:,2).^3) +currPrices(:,1).*(currPrices(:,2).^4)); end if numFns >= 4 condCF = condCF + b(4)*((currPrices(:,1).^2).*currPrices(:,2) +(currPrices(:,1).^2).*(currPrices(:,2).^2) ... + (currPrices(:,1).^2).*(currPrices(:,2).^3)); end if numFns >= 5 condCF = condCF + b(5)*((currPrices(:,1).^3).*currPrices(:,2) +(currPrices(:,1).^3).*(currPrices(:,2).^2)); end if numFns >= 6 condCF = condCF + b(6)*((currPrices(:,1).^4).*currPrices(:,2)); end if numFns >= 7 condCF = condCF + b(7); endelseif strcmp(fnType,all) condCF = condCF + b(1)*sum(Hermite(currPrices(:,1).^2,numFns),2); condCF = condCF + b(2)*sum(Hermite(currPrices(:,2).^2,numFns),2); condCF = condCF + b(3)*sum(Hermite(currPrices(:,3).^2,numFns),2); condCF = condCF + b(4)*sum(Hermite(currPrices(:,4).^2,numFns),2); condCF = condCF + b(5)*sum(Hermite(currPrices(:,5).^2,numFns),2); condCF = condCF +b(6)*sum(Hermite(currPrices(:,1).*currPrices(:,2),numFns),2); condCF = condCF +b(7)*sum(Hermite((currPrices(:,1).^2).*currPrices(:,2),numFns),2); condCF = condCF +b(8)*sum(Hermite(currPrices(:,1).*currPrices(:,3),numFns),2);
- 40. condCF = condCF +b(9)*sum(Hermite(currPrices(:,1).*currPrices(:,4),numFns),2); condCF = condCF +b(10)*sum(Hermite(currPrices(:,1).*currPrices(:,5),numFns),2); condCF = condCF +b(11)*sum(Hermite(currPrices(:,2).*currPrices(:,3),numFns),2); condCF = condCF +b(12)*sum(Hermite(currPrices(:,1).*currPrices(:,2),numFns),2)... +sum(Hermite(currPrices(:,2).*currPrices(:,3),numFns),2);else condCF = condCF + b(1)*crossTerms(:,1); if crossProducts >= 2 condCF = condCF + b(2)*crossTerms(:,2); end if crossProducts >= 3 condCF = condCF + b(3)*crossTerms(:,3); end if crossProducts >= 4 condCF = condCF + b(4)*crossTerms(:,4); end if crossProducts >= 5 condCF = condCF + b(5)*crossTerms(:,5); end if crossProducts >= 6 condCF = condCF + b(6)*crossTerms(:,6); end if crossProducts >= 7 condCF = condCF + b(7)*crossTerms(:,7); endendcondCF = condCF.*moneyness;endfunction out=laguerre(x, n)y0=ones(size(x,1),1);y1=-x+1;y2=0.5*(x.^2-4*x+2);y3=(1/6)*(-x.^3+9*x.^2-18*x+6);
- 41. y4=(1/24)*(x.^4-16*x.^3+72*x.^2-96*x+24);y5=(1/120)*(-x.^5+25*x.^4-200*x.^3+600*x.^2-600*x+120);y6=(1/720)*(x.^6-36*x.^5+450*x.^4-2400*x.^3+5400*x.^2-4320*x+720);Y=[y0 y1 y2 y3 y4 y5 y6];out=Y(:,1:n);endfunction out=Chebyshev1st(x, n)y0=ones(size(x,1),1);y1=x;y2=2*x.^2-1;y3=4*x.^3-3*x;y4=8*x.^4-8*x.^2+1;y5=16*x.^5-20*x.^3+5*x;y6=32*x.^6-48*x.^4+18*x.^2-1;y7=64*x.^7-112*x.^5+56*x.^3-7*x;y8=128*x.^8-256*x.^6+160*x.^4-32*x.^2+1;y9=256*x.^9-576*x.^7+432*x.^5-120*x.^3+9*x;Y=[y0 y1 y2 y3 y4 y5 y6 y7 y8 y9];out=Y(:,1:n);endfunction out=Chebyshev2nd(x, n)y0=ones(size(x,1),1);y1=2*x;y2=4*x.^2-1;y3=8*x.^3-4*x;y4=16*x.^4-12*x.^2+1;y5=32*x.^5-32*x.^3+6*x;y6=64*x.^6-80*x.^4+24*x.^2-1;y7=128*x.^7-192*x.^5+80*x.^3-8*x;y8=256*x.^8-448*x.^6+240*x.^4-40*x.^2+1;y9=512*x.^9-1024*x.^7+672*x.^5-160*x.^3+10*x;Y=[y0 y1 y2 y3 y4 y5 y6 y7 y8 y9];out=Y(:,1:n);endfunction out=Hermite(x, n)
- 42. y0=ones(size(x,1),1);y1=x;y2=x.^2-1;y3=x.^3-3*x;y4=x.^4-6*x.^2+3;y5=x.^5-10*x.^3+15*x;y6=x.^6-15*x.^4+45*x.^2-15;y7=x.^7-21*x.^5+105*x.^3-105*x;y8=x.^8-28*x.^6+210*x.^4-420*x.^2+105;y9=x.^9-36*x.^7+378*x.^5-1260*x.^3+945*x;Y=[y0 y1 y2 y3 y4 y5 y6 y7 y8 y9];out=Y(:,1:n);end%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Code for dynamic delta%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%clc;clear;format short g;warning(off, stats:regress:RankDefDesignMat);corMatrix= [1.0000, 0.2957, 0.3729, 0.1930, 0.2438; 0.2957, 1.0000, 0.3819, 0.3738, 0.5604; 0.3729, 0.3819, 1.0000, 0.2846, 0.3097; 0.1930, 0.3738, 0.2846, 1.0000, 0.2846; 0.2438, 0.5604, 0.3097, 0.2846, 1.0000];corMatrix = eye(1);vols=[.15;.2;.23;0.3;0.1];divYields=[0.1;0.1;0.1;0.1;0.1];spots=[250;40;23;60;3.8];strikes=[250;40;23;60;3.8];barriers=spots*1.2;cycles = 1000;size = 1;foreignRate = 0.05;obsFreq = 16;basketSize = 5;extraAssets = 0;
- 43. rebate = 0.3;domesticRate = 0.01;fxVol = 0.15;fxRate = 100;maturity = 1;batches = 1;basisType = Hermite;basisFns = 10;crossProducts = 3;dimensions=(basketSize + extraAssets + 1)*(obsFreq*maturity);useBrownianBridge = 1;productType = bermudan;% parametersinterval = 1/obsFreq;payoffs = zeros(batches,1);cholMatrix=chol(corMatrix);time_steps = obsFreq*maturity + 1;dimensions=(basketSize + extraAssets + 1)*(time_steps-1);pertDS = 0.0001; %this is the percentage perturbation in the underlyingassetdummy = 0;batches_hedge = 100;termValues = [];cumcostYen_dump=zeros(batches_hedge,time_steps-1);der_price=zeros(batches_hedge,time_steps-1);num_TotIntervals = time_steps; %number of total intevals over the entirematurityfor q = 1:batches_hedgecurrentPort= zeros(1,basketSize+extraAssets+2);%generate the realized trajectories to carry out the hedging simulationrealTraj=(gen_traj(maturity, time_steps-1,corMatrix, vols,... fxVol, spots,fxRate, divYields,domesticRate,foreignRate,num_TotIntervals));cumcostYen = 0; %the cumulative cost in Japanese yendummy = 0; %dummy variable indicates if the barrier has been breached for i = 1:time_steps-1 if i~=1 costYen=0;
- 44. cumcostYen=cumcostYen*exp(domesticRate*1/obsFreq);%interest accrual in JPYcurrentPort(1,end-1:end)=[currentPort(1,end-1)*exp(foreignRate*1/obsFreq),... currentPort(1,end)*exp(domesticRate*1/obsFreq)]; %interestaccrual in USD oldMMUSD = currentPort(1,6); %the USD money market accountprior to delta hedging %if barrier is breached cash out the position if max(realTraj(i,1:5))>barriers terminalValue =(currentPort(1,1:end-2)*realTraj(i,1:end-1)...+currentPort(1,end-1))*realTraj(i,end)+currentPort(1,end); dummy = 1; end oldStockUSD=currentPort(1:end-2)*realTraj(i,1:end-1);%the USD value of the stocks currentPort(1:end-2) = delta; %update the current portfolioholding to delta newStockUSD=currentPort(1:end-2)*realTraj(i,1:end-1); %newportfolio holdingscurrentPort(1,6)=currentPort(1,6)+(oldStockUSD-newStockUSD); %updatethe USD money market newMMUSD = currentPort(1,6); %update the USD money marketaccount after delta hedging currentPort(1,7)=currentPort(1,7)+(oldMMUSD -newMMUSD)*realTraj(i,end); %convert USD money market to JPY costYen=(oldMMUSD - newMMUSD)*realTraj(i,end); %record thecurrent cost of hedging in JPY cumcostYen=cumcostYen+costYen; %update the cumulative cost inJPY cumcostYen_dump(q,i-1)=cumcostYen; %store the values of thecumcostYen end if dummy ~= 1 norm_spots=realTraj(i,1:5)./realTraj(i,1:5); %normalize thespots to 1
- 45. %select the price at time step i and replicate trajectories = repmat(norm_spots,cycles,1); fxTrajectories = ones(cycles,time_steps); %reshape the 2D matrix into a 3D matrix temp =reshape(trajectories,cycles,1,basketSize+extraAssets); trajectories=zeros(cycles, time_steps+1-i, basketSize +extraAssets); trajectoriesP=zeros(cycles, time_steps+1-i, basketSize +extraAssets); trajectories(:,1,:) = temp; %create a matrix to store the brownian motions to use the same %brownian motion for the perturbed and unperturbed trajectories zMat =zeros(cycles, time_steps-i, basketSize + extraAssets); fxZMat =zeros(cycles, time_steps-i); for p=1:time_steps-i zslice=(cholMatrix*normrnd(0,1,basketSize +extraAssets,cycles)); zslice=reshape(zslice,[cycles,1,basketSize +extraAssets]); zMat(:,p,:)=zslice; fxZ=randn(cycles,time_steps-i); end for j=2:time_steps-i+1 % generate unperturbed trajectories for k=1:basketSize + extraAssets trajectories(:,j,k) =trajectories(:,j-1,k).*exp((foreignRate... -divYields(k)-0.5*(vols(k)^2))*interval +vols(k)*sqrt(interval)*zMat(:,j-1,k)); end fxTrajectories(:,j) =fxTrajectories(:,j-1).*exp((domesticRate... -foreignRate-0.5*(fxVol^2))*interval +fxVol*sqrt(interval)*fxZ(:,j-1)); end % generate perturbed trajectories by perturbing at time step 1 for k=1:basketSize + extraAssets
- 46. trajectoriesP(:,1,k) = trajectories(:,1,k)*(1+pertDS); end fxTrajectoriesP(:,1) = fxTrajectories(:,1)*(1+pertDS); %let the perturbed trajectories propagate from time step 2through end for each asset for j=2:time_steps-i+1 % generate trajectories for k=1:basketSize + extraAssets trajectoriesP(:,j,k) =trajectoriesP(:,j-1,k).*exp((foreignRate... -divYields(k)-0.5*(vols(k)^2))*interval +vols(k)*sqrt(interval)*zMat(:,j-1,k)); end fxTrajectoriesP(:,j) =fxTrajectoriesP(:,j-1).*exp((domesticRate... -foreignRate-0.5*(fxVol^2))*interval +fxVol*sqrt(interval)*fxZ(:,j-1)); end %Assign the unperturbed paths paths = trajectories; fxpaths = fxTrajectories; %output the price of the portfolio prior to perturbation nMC_results = jrd_LSMC(maturity-(i-1)/obsFreq, vols,divYields, foreignRate,... spots, strikes, obsFreq+1-i, cycles, corMatrix,basketSize,extraAssets, ... barriers, rebate, domesticRate, fxVol,fxRate,batches,naive,useBrownianBridge,...basisType,basisFns,crossProducts,bermudan,1,paths,fxpaths); der_price(q,i)=(nMC_results(1)); results =[]; pricesPert = []; pathsPert=zeros(cycles, time_steps+1-i, basketSize +extraAssets); %Assign the perturbed paths for each asset and for the fx for k=1:basketSize + extraAssets pathsPert = trajectories; pathsPert(:,:,k) = trajectoriesP(:,:,k);
- 47. %output the price of the portfolio after the perturbation nMC_resultsP = jrd_LSMC(maturity-(i-1)/obsFreq, vols,divYields, foreignRate,... spots, strikes, obsFreq+1-i, cycles, corMatrix,basketSize,extraAssets, ... barriers, rebate, domesticRate, fxVol,fxRate,batches,naive,useBrownianBridge,...basisType,basisFns,crossProducts,bermudan,1,pathsPert,fxpaths); results = [results nMC_resultsP(1)]; pricesPert = [pricesPerttrajectories(1,1,k)*pertDS*realTraj(i,k)]; end delta = (results - nMC_results(1))./pricesPert fxpathP = fxTrajectoriesP; %output the price of the portfolio after perturbing the fx rate nMC_resultsPfx = jrd_LSMC(maturity-(i-1)/obsFreq, vols,divYields, foreignRate,... spots, strikes, obsFreq+1-i, cycles, corMatrix,basketSize,extraAssets, ... barriers, rebate, domesticRate, fxVol,fxRate,batches,naive,useBrownianBridge,...basisType,basisFns,crossProducts,bermudan,1,paths,fxpathP); FXpricePert = fxTrajectories(1,i)*pertDS; deltaFX = (nMC_resultsPfx(1) - nMC_results(1))/FXpricePert; else break end end %out the terminal value of the hedging portfolio terminalValue = (currentPort(1,1:end-2)*realTraj(end,1:end-1)...+currentPort(1,end-1))*realTraj(end,end)+currentPort(1,end); termValues = [termValues terminalValue];endxlswrite(deltaHedging.xls,termValues,Terminal Values);hist(termValues,20);%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Program to generate exercise boundaries
- 48. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%boundaryplot=[];corMatrix= [1.0000, 0.2957, 0.3729, 0.1930, 0.2438; 0.2957, 1.0000, 0.3819, 0.3738, 0.5604; 0.3729, 0.3819, 1.0000, 0.2846, 0.3097; 0.1930, 0.3738, 0.2846, 1.0000, 0.2846; 0.2438, 0.5604, 0.3097, 0.2846, 1.0000];vols=[.15;.2;.23;0.3;0.1];divYields=[0.1;0.1;0.1;0.1;0.1];spots=[250;40;23;60;3.8];strikes=[250;40;23;60;3.8];barMultiplier=[1.2:0.1:2];cycles = 100000;size = 1;foreignRate = 0.05;observation_frequency = 24;basketSize = 5;extraAssets = 0;rebate = 0.3;domesticRate = 0.01;fxVol = 0.15;fxRate = 100;maturity = 1;batches = 1;for counter=1:length(barMultiplier)% barriers=spots*barMultiplier(counter);if counter == 1 color = b; basisType = Hermite;elseif counter == 2 color = g; basisType = monomial;elseif counter == 3 color = r; basisType = laguerre;end
- 49. basisFns = 10;crossProducts = 3;dimensions=(basketSize + extraAssets +1)*(observation_frequency*maturity);useBrownianBridge = 0;productType = bermudan;bound=boundaryCal(maturity, vols, divYields, foreignRate,...spots, strikes,observation_frequency, cycles, corMatrix,basketSize,extraAssets, ... barriers, rebate, domesticRate, fxVol, fxRate);disp(bound);plot(bound,color);boundaryplot=[boundaryplot; bound];hold onendlegend(Hermite,monomial,laguerre);legend(gca,boxoff);xlabel(Time);ylabel(Normalized stock Price);function out=boundaryCal(maturity, vols, divYields, foreignRate,...spots, strikes,observation_frequency, cycles, corMatrix,basketSize,extraAssets, ... highBarrier, rebate, domesticRate, fxVol, fxRate)% parametersinterval = 1/observation_frequency;cholMatrix=chol(corMatrix);strikes = strikes./spots; %scale strike for ease of use in basis functionshighBarrier= highBarrier./spots;%for i=1:batches % work with normalizes stock prices and fx rates, hence period 0 values % are 1 trajectories = ones(cycles, observation_frequency*maturity+1,basketSize + extraAssets); fxTrajectories = ones(cycles,observation_frequency*maturity+1);
- 50. for j=2:observation_frequency*maturity+1 z=(cholMatrix*normrnd(0,1,basketSize + extraAssets,cycles)); for k=1:basketSize + extraAssets trajectories(:,j,k) =trajectories(:,j-1,k).*exp((foreignRate... -divYields(k)-0.5*(vols(k)^2))*interval +vols(k)*sqrt(interval)*z(:,k)); end fxTrajectories(:,j) =fxTrajectories(:,j-1).*exp((domesticRate... -foreignRate-0.5*(fxVol^2))*interval + fxVol* ... sqrt(interval)*randn(cycles,1)); end clear z cholMatrix; % free up memory % work backwards cfMatrix = zeros(cycles, observation_frequency*maturity); cfMatrixPlusRabate= zeros(cycles, observation_frequency*maturity); stillAlive=zeros(cycles, observation_frequency*maturity); % determine final payoff based on initial composition of the basket currPrice =reshape(trajectories(:,observation_frequency*maturity+1,... 1:basketSize),cycles, basketSize); cfMatrix(:,observation_frequency*maturity) = (max(max(currPrice... -repmat(strikes(1:basketSize),cycles,1),0),[],2)).*fxTrajectories(:,observation_frequency*maturity); alive=ones(cycles,1); for m=1:observation_frequency*maturity temp=reshape(trajectories(:,m+1,:),cycles,basketSize); alive=alive.* min((repmat(highBarrier,cycles,1)-temp)>0,[],2); stillAlive(:,m)=alive; end breachingPoint=[ones(cycles,1),stillAlive(:,1:(end-1))]-stillAlive; cfMatrix(:,observation_frequency*maturity)=cfMatrix(:,observation_frequency*maturity).*stillAlive(:,observation_frequency*maturity); cfMatrixPlusRabate(:,observation_frequency*maturity)=
- 51. cfMatrix(:,observation_frequency*maturity)+rebate.*breachingPoint(:,observation_frequency*maturity); for j=observation_frequency*maturity-1:-1:1 % determing current exercise value currPrice = reshape(trajectories(:,j+1,1:basketSize),cycles,basketSize); exerciseCF = (max(max(currPrice -repmat(strikes(1:basketSize),cycles,1),0),[],2)).*fxTrajectories(:,j+1); exerciseCF = exerciseCF.*stillAlive(:,j); % cannot exercise ifknocked out moneyness = exerciseCF>0; pvVector = zeros(cycles,1); for k=j+1:observation_frequency*maturity % As all future cashflows have already been converted to the % domestic currency discounting done in domestic rate pvVector = pvVector +cfMatrixPlusRabate(:,k)*exp(-domesticRate*interval*(k-j)); end % only consider the two best performing assets for the least % squares regression sortedPrices =reshape(sort(trajectories(:,j+1,1:basketSize),3,descend),cycles,basketSize); currPrices = sortedPrices(:,1:2); condExpect = LSR_mono(currPrices,pvVector,3,7,moneyness); cfMatrix(:,j) = moneyness.*((condExpect < exerciseCF).*exerciseCF); % If knocked out pay the rebate cfMatrixPlusRabate(:,j)=cfMatrix(:,j)+rebate*breachingPoint(:,j); % set CFs of j+1 to maturity zero if exercised at current time for k=j+1:observation_frequency*maturity % If not exercised and not knocked out, keep old cashflows cfMatrix(:,k) = cfMatrix(:,k).*(cfMatrix(:,j)==0); cfMatrixPlusRabate(:,k) =cfMatrixPlusRabate(:,k).*(cfMatrix(:,j)==0) end end temp=ones(cycles, observation_frequency*maturity);
- 52. temp=max(trajectories(:,2:end,:),[],3); temp1=temp.*(cfMatrix~=0); index = find(temp1 == 0); temp(index)=NaN; boundary=nanmin(temp,[],1);clear trajectories fxTrajectories; % free memoryout=boundary;end%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Code for static hedging%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%hedging=staticH(maturity, vols, divYields, foreignRate,...spots, strikes,observation_frequency, cycles, corMatrix,basketSize,extraAssets, ... barriers, rebate, domesticRate, fxVol, fxRate);% regression[b,bint,residuals]=regress(hedging(2:end,1),hedging(2:end,2:end));% P&LhedgingPL=-residuals+(hedging(1,1)-(b(1)*hedging(1,2)+b(2)*hedging(1,3)+b(3)*hedging(1,4)));noHedgingPL=mean(hedging(:,1))-hedging(:,1);% Plot and summaryfigure(1);hist(hedgingPL,100);xlabel(P&L);ylabel(frequency);xlim([-0.6 0.6]);ylim([0 2000]);title(With MMC Hedging)quantile(hedgingPL,[0.01,0.05,0.25])figure(2);hist(noHedgingPL,100);xlabel(P&L);