def test_FinEquityBasketOption():

    import time

    valueDate = FinDate(2015, 1, 1)
    expiryDate = FinDate(2016, 1, 1)
    volatility = 0.30
    interestRate = 0.05
    discountCurve = FinDiscountCurveFlat(valueDate, interestRate)

    ##########################################################################
    # Homogeneous Basket
    ##########################################################################

    numAssets = 5
    volatilities = np.ones(numAssets) * volatility
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100

    betaList = np.linspace(0.0, 0.999999, 11)

    testCases.header("NumPaths", "Beta", "Value", "ValueMC", "TIME")

    for beta in betaList:
        for numPaths in [10000]:
            callOption = FinEquityBasketOption(expiryDate, 100.0,
                                               FinOptionTypes.EUROPEAN_CALL,
                                               numAssets)
            betas = np.ones(numAssets) * beta
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()
            v = callOption.value(valueDate, stockPrices, discountCurve,
                                 dividendYields, volatilities, corrMatrix)

            vMC = callOption.valueMC(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix,
                                     numPaths)
            end = time.time()
            duration = end - start
            testCases.print(numPaths, beta, v, vMC, duration)

    ##########################################################################
    # INHomogeneous Basket
    ##########################################################################

    numAssets = 5
    volatilities = np.array([0.3, 0.2, 0.25, 0.22, 0.4])
    dividendYields = np.array([0.01, 0.02, 0.04, 0.01, 0.02])
    stockPrices = np.array([100, 105, 120, 100, 90])

    betaList = np.linspace(0.0, 0.999999, 11)

    testCases.header("NumPaths", "Beta", "Value", "ValueMC", "TIME")

    for beta in betaList:

        for numPaths in [10000]:

            callOption = FinEquityBasketOption(expiryDate, 100.0,
                                               FinOptionTypes.EUROPEAN_CALL,
                                               numAssets)
            betas = np.ones(numAssets) * beta
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()

            v = callOption.value(valueDate, stockPrices, discountCurve,
                                 dividendYields, volatilities, corrMatrix)

            vMC = callOption.valueMC(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix,
                                     numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, beta, v, vMC, duration)

    ##########################################################################
    # Homogeneous Basket
    ##########################################################################

    numAssets = 5
    volatilities = np.ones(numAssets) * volatility
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100
    betaList = np.linspace(0.0, 0.999999, 11)

    testCases.header("NumPaths", "Beta", "Value", "ValueMC", "TIME")

    for beta in betaList:
        for numPaths in [10000]:
            callOption = FinEquityBasketOption(expiryDate, 100.0,
                                               FinOptionTypes.EUROPEAN_PUT,
                                               numAssets)
            betas = np.ones(numAssets) * beta
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()
            v = callOption.value(valueDate, stockPrices, discountCurve,
                                 dividendYields, volatilities, corrMatrix)
            vMC = callOption.valueMC(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix,
                                     numPaths)
            end = time.time()
            duration = end - start
            testCases.print(numPaths, beta, v, vMC, duration)

    ##########################################################################
    # INHomogeneous Basket
    ##########################################################################

    numAssets = 5
    volatilities = np.array([0.3, 0.2, 0.25, 0.22, 0.4])
    dividendYields = np.array([0.01, 0.02, 0.04, 0.01, 0.02])
    stockPrices = np.array([100, 105, 120, 100, 90])
    betaList = np.linspace(0.0, 0.999999, 11)

    testCases.header("NumPaths", "Beta", "Value", "ValueMC", "TIME")

    for beta in betaList:

        for numPaths in [10000]:

            callOption = FinEquityBasketOption(expiryDate, 100.0,
                                               FinOptionTypes.EUROPEAN_PUT,
                                               numAssets)
            betas = np.ones(numAssets) * beta
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()
            v = callOption.value(valueDate, stockPrices, discountCurve,
                                 dividendYields, volatilities, corrMatrix)
            vMC = callOption.valueMC(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix,
                                     numPaths)
            end = time.time()
            duration = end - start
            testCases.print(numPaths, beta, v, vMC, duration)
Beispiel #2
0
def test_FinEquityRainbowOption():

    #        import matplotlib.pyplot as plt

    valueDate = FinDate(2015, 1, 1)
    expiryDate = FinDate(2016, 1, 1)
    interestRate = 0.05
    discountCurve = FinDiscountCurveFlat(valueDate, interestRate)

    numAssets = 2
    volatilities = np.ones(numAssets) * 0.3
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100
    numPathsList = [10000]
    corrList = np.linspace(0.0, 0.999999, 6)
    strike = 100.0

    testCases.banner(
        "===================================================================")
    testCases.banner("                      CALL ON MAXIMUM")
    testCases.banner(
        "===================================================================")

    payoffType = FinEquityRainbowOptionTypes.CALL_ON_MAXIMUM
    payoffParams = [strike]
    rainbowOption = FinEquityRainbowOption(expiryDate, payoffType,
                                           payoffParams, numAssets)

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            start = time.time()
            v = rainbowOption.value(valueDate, stockPrices, discountCurve,
                                    dividendYields, volatilities, corrMatrix)

            v_MC = rainbowOption.valueMC(valueDate, stockPrices, discountCurve,
                                         dividendYields, volatilities,
                                         corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "CALL ON MAX Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "CALL ON MAX Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

##########################################################################

    testCases.banner(
        "===================================================================")
    testCases.banner("                       CALL ON MINIMUM")
    testCases.banner(
        "===================================================================")
    payoffType = FinEquityRainbowOptionTypes.CALL_ON_MINIMUM
    payoffParams = [strike]
    rainbowOption = FinEquityRainbowOption(expiryDate, payoffType,
                                           payoffParams, numAssets)

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            start = time.time()

            v = rainbowOption.value(valueDate, stockPrices, discountCurve,
                                    dividendYields, volatilities, corrMatrix)

            v_MC = rainbowOption.valueMC(valueDate, stockPrices, discountCurve,
                                         dividendYields, volatilities,
                                         corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "CALL ON MIN Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "CALL ON MIN Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

###############################################################################

    testCases.banner(
        "===================================================================")
    testCases.banner("                      PUT ON MAXIMUM")
    testCases.banner(
        "===================================================================")

    payoffType = FinEquityRainbowOptionTypes.PUT_ON_MAXIMUM
    payoffParams = [strike]
    rainbowOption = FinEquityRainbowOption(expiryDate, payoffType,
                                           payoffParams, numAssets)

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            start = time.time()

            v = rainbowOption.value(valueDate, stockPrices, discountCurve,
                                    dividendYields, volatilities, corrMatrix)

            v_MC = rainbowOption.valueMC(valueDate, stockPrices, discountCurve,
                                         dividendYields, volatilities,
                                         corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "PUT ON MAX Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "PUT ON MAX Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

##########################################################################

    testCases.banner(
        "===================================================================")
    testCases.banner("                       PUT ON MINIMUM")
    testCases.banner(
        "===================================================================")
    payoffType = FinEquityRainbowOptionTypes.PUT_ON_MINIMUM
    payoffParams = [strike]
    rainbowOption = FinEquityRainbowOption(expiryDate, payoffType,
                                           payoffParams, numAssets)

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            start = time.time()
            v = rainbowOption.value(valueDate, stockPrices, discountCurve,
                                    dividendYields, volatilities, corrMatrix)
            v_MC = rainbowOption.valueMC(valueDate, stockPrices, discountCurve,
                                         dividendYields, volatilities,
                                         corrMatrix, numPaths)
            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "PUT ON MIN Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "PUT ON MIN Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

##########################################################################

    numAssets = 2
    volatilities = np.ones(numAssets) * 0.3
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100
    strike = 100.0
    correlation = 0.50

    testCases.banner(
        "===================================================================")
    testCases.banner("                      CALL ON 1st")
    testCases.banner(
        "===================================================================")

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            payoffType1 = FinEquityRainbowOptionTypes.CALL_ON_MAXIMUM
            payoffParams1 = [strike]
            rainbowOption1 = FinEquityRainbowOption(expiryDate, payoffType1,
                                                    payoffParams1, numAssets)

            payoffType2 = FinEquityRainbowOptionTypes.CALL_ON_NTH
            payoffParams2 = [1, strike]
            rainbowOption2 = FinEquityRainbowOption(expiryDate, payoffType2,
                                                    payoffParams2, numAssets)

            start = time.time()

            v = rainbowOption1.value(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix)

            v_MC = rainbowOption2.valueMC(valueDate, stockPrices,
                                          discountCurve, dividendYields,
                                          volatilities, corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "CALL ON MAX Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "CALL ON 1st Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

    testCases.banner(
        "===================================================================")
    testCases.banner("                      CALL ON 2nd")
    testCases.banner(
        "===================================================================")

    rainboxOptionValues = []
    rainbowOptionValuesMC = []

    testCases.header("NUMPATHS", "CORRELATION", "VALUE", "VALUE_MC", "TIME")

    for correlation in corrList:

        betas = np.ones(numAssets) * sqrt(correlation)
        corrMatrix = betaVectorToCorrMatrix(betas)

        for numPaths in numPathsList:

            payoffType1 = FinEquityRainbowOptionTypes.CALL_ON_MINIMUM
            payoffParams1 = [strike]
            rainbowOption1 = FinEquityRainbowOption(expiryDate, payoffType1,
                                                    payoffParams1, numAssets)

            payoffType2 = FinEquityRainbowOptionTypes.CALL_ON_NTH
            payoffParams2 = [2, strike]
            rainbowOption2 = FinEquityRainbowOption(expiryDate, payoffType2,
                                                    payoffParams2, numAssets)

            start = time.time()

            v = rainbowOption1.value(valueDate, stockPrices, discountCurve,
                                     dividendYields, volatilities, corrMatrix)

            v_MC = rainbowOption2.valueMC(valueDate, stockPrices,
                                          discountCurve, dividendYields,
                                          volatilities, corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, v, v_MC, duration)

            rainboxOptionValues.append(v)
            rainbowOptionValuesMC.append(v_MC)

#    plt.figure(figsize=(10,8))
#    plt.plot(corrList, rainboxOptionValues, color = 'r', label = "CALL ON MIN Rainbow Option Analytical")
#    plt.plot(corrList, rainbowOptionValuesMC, 'o', color = 'b', label = "CALL ON 2nd Rainbow Option MC")
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

    testCases.banner(
        "===================================================================")
    testCases.banner("                      CALL ON 1-5")
    testCases.banner(
        "===================================================================")

    rainboxOptionValues = []
    rainbowOptionValuesMC = []
    numPaths = 10000
    numAssets = 5
    volatilities = np.ones(numAssets) * 0.3
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100

    #    plt.figure(figsize=(10,8))

    testCases.header("NUMPATHS", "CORRELATION", "NTD", "VALUE", "VALUE_MC",
                     "TIME")

    for n in [1, 2, 3, 4, 5]:

        rainboxOptionValues = []
        rainbowOptionValuesMC = []

        payoffType2 = FinEquityRainbowOptionTypes.CALL_ON_NTH
        payoffParams2 = [n, strike]
        rainbowOption2 = FinEquityRainbowOption(expiryDate, payoffType2,
                                                payoffParams2, numAssets)

        for correlation in corrList:

            betas = np.ones(numAssets) * sqrt(correlation)
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()

            v_MC = rainbowOption2.valueMC(valueDate, stockPrices,
                                          discountCurve, dividendYields,
                                          volatilities, corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, n, v, v_MC, duration)

            rainbowOptionValuesMC.append(v_MC)

#        plt.plot(corrList, rainbowOptionValuesMC, 'o-', label = "CALL Rainbow Option MC NTH = " + str(n))
#    plt.xlabel("Correlation")
#    plt.legend(loc='best')

    testCases.banner(
        "===================================================================")
    testCases.banner("                      PUT ON 1-5")
    testCases.banner(
        "===================================================================")

    rainboxOptionValues = []
    rainbowOptionValuesMC = []
    numPaths = 10000
    numAssets = 5
    volatilities = np.ones(numAssets) * 0.3
    dividendYields = np.ones(numAssets) * 0.01
    stockPrices = np.ones(numAssets) * 100

    #    plt.figure(figsize=(10,8))

    testCases.header("NUMPATHS", "CORRELATION", "NTD", "VALUE", "VALUE_MC",
                     "TIME")

    for n in [1, 2, 3, 4, 5]:

        rainboxOptionValues = []
        rainbowOptionValuesMC = []

        payoffType2 = FinEquityRainbowOptionTypes.PUT_ON_NTH
        payoffParams2 = [n, strike]
        rainbowOption2 = FinEquityRainbowOption(expiryDate, payoffType2,
                                                payoffParams2, numAssets)

        for correlation in corrList:

            betas = np.ones(numAssets) * sqrt(correlation)
            corrMatrix = betaVectorToCorrMatrix(betas)

            start = time.time()

            v_MC = rainbowOption2.valueMC(valueDate, stockPrices,
                                          discountCurve, dividendYields,
                                          volatilities, corrMatrix, numPaths)

            end = time.time()
            duration = end - start
            testCases.print(numPaths, correlation, n, v, v_MC, duration)

            rainbowOptionValuesMC.append(v_MC)