Python金融大数据分析——第16章 金融模型的模拟 笔记

第16章 金融模型的模拟

本章会用到 金融数据分析库 DX
下载地址：https://github.com/yhilpisch/dx
找到setup.py，手动安装：python setup.py install
我们要用到的很多模型这个库中都有

参考：Python金融大数据分析——第10章 推断统计学

16.1 随机数生成

生成标准正态分布随机数的函数

# 生成标准正态分布随机数的函数
import numpy as np


def sn_random_numbers(shape, antithetic=True, moment_matching=True, fixed_seed=False):
    """
    Returns an array of shape shape with (pseudo) random numbers
    that are standard normally distributed.
    :param shape: tuple(0,n,m)
        generation of array with shape(0,n,m)
    :param antithetic: Boolean
        generation of antithetic variates
    :param moment_matching: Boolean
        matching of first and second moments
    :param fixed_seed: Boolean
        flag to fix and seed
    :return:
        ran:(0,n,m) array of (pseudo) random numbers
    """
    if fixed_seed:
        np.random.seed(1000)

    if antithetic:
        ran = np.random.standard_normal((shape[0], shape[1], shape[2] / 2))
        ran = np.concatenate((ran, -ran), axis=2)
    else:
        ran = np.random.standard_normal(shape)

    if moment_matching:
        ran = ran - np.mean(ran)
        ran = ran / np.std(ran)

    if shape[0] == 1:
        return ran[0]
    else:
        return ran


snrn = sn_random_numbers((2, 2, 2), antithetic=False, moment_matching=False, fixed_seed=True)
snrn
# array([[[-0.8044583 ,  0.32093155],
#         [-0.02548288,  0.64432383]],
#        [[-0.30079667,  0.38947455],
#         [-0.1074373 , -0.47998308]]])

snrn_mm = sn_random_numbers((2, 3, 2), antithetic=False, moment_matching=True, fixed_seed=True)
snrn_mm
# array([[[-1.47414161,  0.67072537],
#         [ 0.01049828,  1.28707482],
#         [-0.51421897,  0.80136066]],
#        [[-0.14569767, -0.85572818],
#         [ 1.19313679, -0.82653845],
#         [ 1.3308292 , -1.47730025]]])

snrn_mm.mean()
# 3.700743415417188e-17
snrn_mm.std()
# 1.0

16.2 泛型模拟类

面向对象建模允许属性和方法的继承。这是我们在构建模拟类时想要利用的：从一个泛型模拟类开始，它包含所有其他模拟类共享的属性和方法。

首先要注意的是， 我们实例化任何模拟类的一个对象都 “ 只” 提供3个属性：

name：用作模型模拟对象名称的字符科才象

mar_env ： maket_environment 类的一个实例

corr：表示对象是否相关的一个标志（布尔型）

这再次说明了市场环境的作用：在一步中提供模拟和估值所需的所有数据和对象。泛型类的方法如下：

generate_time_grid：
这个方法生成模拟所用的相关日期的 “时间网格”。这个任务对于每个模拟类都相同。

get_instument_values：
每个模拟类都必须返回包含模拟金融工具价值的 ndarray 对象（例如模拟的股票价格， 商品价格， 波动率）。

泛型金融模型模拟类

# 泛型金融模型模拟类
import numpy as np
import pandas as pd


class simulation_class(object):
    """
    Providing base methods for simulation classes.
    """

    def __init__(self, name, mar_env, corr):
        """
        :param name: string
            name of the object
        :param mar_env: instance of market_environment
            market environment data for simulation
        :param corr: Boolean
            True of correlated with other model object
        """
        try:
            self.name = name
            self.pricing_data = mar_env.pricing_date
            self.initial_value = mar_env.get_constant('initial_value')
            self.volatility = mar_env.get_constant('volatility')
            self.final_date = mar_env.get_constant('final_date')
            self.currency = mar_env.get_constant('currency')
            self.frequency = mar_env.get_constant('frequency')
            self.paths = mar_env.get_constant('paths')
            self.discount_curve = mar_env.get_curve('discount_curve')
            try:
                # if time_grid in mar_env take this
                # (for portfolio valuation)
                self.time_grid = mar_env.get_list('time_grid')
            except:
                self.time_grid = None
            try:
                # if there are special dates, then add these
                self.special_dates = mar_env.get_list('special_dates')
            except:
                self.special_dates = []
            self.instrument_values = None
            self.correlated = corr
            if corr:
                # only needed in a portfolio context when
                # risk factors are correlated
                self.cholesky_matrix = mar_env.get_list('cholesky_matrix')
                self.rn_set = mar_env.get_list('rn_set')[self.name]
                self.random_numbers = mar_env.get_list('random_numbers')
        except:
            print("Error parsing market environment.")

    def generate_time_grid(self):
        start = self.pricing_data
        end = self.final_date
        # pandas date_range function
        # freq = e.g. 'B' for Business Day,
        # 'W' for Weekly, 'M'for Monthly
        time_grid = pd.date_range(start=start, end=end, freq=self.frequency).to_pydatetime()
        time_grid = list(time_grid)
        if start not in time_grid:
            time_grid.insert(0, start)  # insert start date if not in list
        if end not in time_grid:
            time_grid.append(end)  # insert end date if not in list
        if len(self.special_dates) > 0:
            # add all special dates
            time_grid.extend(self.special_dates)
            # delete duplicates
            time_grid = list(set(time_grid))
            # sort list
            time_grid.sort()
        self.time_grid = np.array(time_grid)

    def get_instrument_values(self, fixed_seed=True):
        if self.instrument_values is None:
            # only initiate simulation if there are no instrument values
            self.generate_paths(fixed_seed=fixed_seed, day_count=365.)
        elif fixed_seed is False:
            # also initiate resimulation when fixed_seed is False
            self.generate_paths(fixed_seed=fixed_seed, day_count=365.)
        return self.instrument_values

所有模拟类的市场环境元素

元素	类型	强制	描述
initial_value	常量	是	pricing_date（定价日）时的过程初始值
volatility	常量	是	过程的波动性系数
final_date	常量	是	模拟范围
cuπency	常量	是	金融实体的货币
fequency	常量	是	日期频率，和pandas freq参数相同
paths	常量	是	模拟路径数量
discount_curve	曲线	是	constant_short_rate 实例
time_grid	列表	否	相关日期的时间网格（在投资组合背景下）
random_numbers	列表	否	随机数数组（用于相关对象）
cholesky_matrix	列表	否	Cholesky 矩阵（用于相关对象）
rn_set	列表	否	包含指向相关随机数值指针的字典对象

16.3 几何布朗运动

公式 几何布朗运动的随机微分方程

dSt=rStdt+σStdZt d S t = r S t d t + σ S t d Z t

下面公式提供了上述微分方程用于模拟目的的欧拉离散化格式。我们工作于离散时间市场模型中，使用有限相关日期集合 0<t1<t2<...<T 0 < t 1 < t 2 < . . . < T 。

公式 模拟几何布朗运动的微分方程

Stm+1=Stmexp((r−12σ2)(tm+1−tm)+σtm+1−tm−−−−−−−−√Zt)0≤tm≤tm+1≤T S t m + 1 = S t m e x p ( ( r − 1 2 σ 2 ) ( t m + 1 − t m ) + σ t m + 1 − t m Z t ) 0 ≤ t m ≤ t m + 1 ≤ T

16.3.1 模拟类

import numpy as np


class geometric_brownian_motion(simulation_class):
    """
    Class to generate simulated paths based on
    the Black_Scholes-Merton geometric Brownian motion model.
    """

    def __init__(self, name, mar_env, corr=False):
        super(geometric_brownian_motion, self).__init__(name, mar_env, corr)

    def update(self, initial_value=None, volatility=None, final_date=None):
        if initial_value is not None:
            self.initial_value = initial_value
        if volatility is not None:
            self.volatility = volatility
        if final_date is not None:
            self.final_date = final_date
        self.instrument_values = None

    def generate_paths(self, fixed_seed=False, day_count=365.):
        if self.time_grid is None:
            self.generate_time_grid()
        M = len(self.time_grid)
        I = self.paths
        paths = np.zeros((M, I))
        paths[0] = self.initial_value
        if not self.correlated:
            rand = sn_random_numbers((1, M, I), fixed_seed=fixed_seed)
        else:
            rand = self.random_numbers
        short_rate = self.discount_curve.short_rate
        for t in range(1, len(self.time_grid)):
            if not self.correlated:
                ran = rand[t]
            else:
                ran = np.dot(self.cholesky_matrix, rand[:, t, :])
                rand = ran[self.rn_set]
            dt = (self.time_grid[t] - self.time_grid[t - 1]).days / day_count
            paths[t] = paths[t - 1] * np.exp((short_rate - 0.5 * self.volatility ** 2) * dt
                                             + self.volatility * np.sqrt(dt) * ran)
        self.instrument_values = paths

16.3.2 用例

from dx import *

me_gbm = market_environment('me_gbm', dt.datetime(2018, 1, 1))
me_gbm.add_constant('initial_value', 36.)
me_gbm.add_constant('volatility', 0.2)
me_gbm.add_constant('final_date', dt.datetime(2018, 12, 31))
me_gbm.add_constant('currency', 'EUR')
me_gbm.add_constant('frequency', 'M')
me_gbm.add_constant('paths', 10000)
csr = constant_short_rate('csr', 0.05)
me_gbm.add_curve('discount_curve', csr)

gbm = geometric_brownian_motion('gbm', me_gbm)
gbm.generate_time_grid()
gbm.time_grid
# array([datetime.datetime(2018, 1, 1, 0, 0),
#        datetime.datetime(2018, 1, 31, 0, 0),
#        datetime.datetime(2018, 2, 28, 0, 0),
#        datetime.datetime(2018, 3, 31, 0, 0),
#        datetime.datetime(2018, 4, 30, 0, 0),
#        datetime.datetime(2018, 5, 31, 0, 0),
#        datetime.datetime(2018, 6, 30, 0, 0),
#        datetime.datetime(2018, 7, 31, 0, 0),
#        datetime.datetime(2018, 8, 31, 0, 0),
#        datetime.datetime(2018, 9, 30, 0, 0),
#        datetime.datetime(2018, 10, 31, 0, 0),
#        datetime.datetime(2018, 11, 30, 0, 0),
#        datetime.datetime(2018, 12, 31, 0, 0)], dtype=object)

%time paths_1 = gbm.get_instrument_values()
# Wall time: 8.99 ms

paths_1
# array([[36.        , 36.        , 36.        , ..., 36.        ,
#         36.        , 36.        ],
#        [37.37221481, 38.08890977, 34.37156575, ..., 36.22258915,
#         35.05503522, 39.63544014],
#        [39.45866146, 42.18817025, 32.38579992, ..., 34.80319951,
#         33.60600939, 37.62733874],
#        ...,
#        [40.15717404, 33.16701733, 23.32556112, ..., 37.5619937 ,
#         29.89282508, 30.2202427 ],
#        [42.0974104 , 36.59006321, 21.70771374, ..., 35.70950512,
#         30.64670854, 30.45901309],
#        [43.33170027, 37.42993532, 23.8840177 , ..., 35.92624556,
#         27.87720187, 28.77424561]])

gbm.update(volatility=0.5)
%time paths_2 = gbm.get_instrument_values()
# Wall time: 8 ms

# GBM 模拟类中的模拟路程
import matplotlib.pyplot as plt

plt.figure(figsize=(8, 6))
p1 = plt.plot(gbm.time_grid, paths_1[:, :10], 'b')
p2 = plt.plot(gbm.time_grid, paths_2[:, :10], 'r-.')
plt.grid(True)
l1 = plt.legend([p1[0], p2[0]], ['low volatility', 'high volatility'], loc=2)
plt.gca().add_artist(l1)
plt.xticks(rotation=30)

GBM 模拟类中的模拟路程

16.4 跳跃扩散

公式 Meron跳跃扩散模型的随分方程

dSt=(r−rJ)Stdt+σStdZt+JtStdNt d S t = ( r − r J ) S t d t + σ S t d Z t + J t S t d N t

St S t ：t日的指数水平
r：恒定无风险短期利率
rJ≡λ⋅(eμJ+δ2/2−1) r J ≡ λ ⋅ ( e μ J + δ 2 / 2 − 1 ) 维持风险中立性的跳跃漂移校正
σ σ ：S的恒定波动率
Zt Z t ：标准布朗运动
Jt J t ：t日呈……分布的跳跃
log(1+Jt)≈N(log(1+μJ)−δ22,δ2) l o g ( 1 + J t ) ≈ N ( l o g ( 1 + μ J ) − δ 2 2 , δ 2 )
N是标准正态随机变量的累积分布函数
Nt N t ：密度为 λ λ 的泊松分布

下面公式介绍一种用于跳跃扩散的欧拉离散化公式，其中 znt z t n 呈标准正态分布，yt呈密度为λ的泊松分布。

公式 Meron跳跃扩散模型的欧拉离散化

St=St−Δt(e(r−rJ−σ2/2)Δt+σΔt√z1t+(eμJ+δz2t−1)yt) S t = S t − Δ t ( e ( r − r J − σ 2 / 2 ) Δ t + σ Δ t z t 1 + ( e μ J + δ z t 2 − 1 ) y t )

16.4.1 模拟类

#
# DX Analytics
# Base Classes and Model Classes for Simulation
# jump_diffusion.py
#

class jump_diffusion(simulation_class):
    ''' Class to generate simulated paths based on
    the Merton (1976) jump diffusion model.

    Attributes
    ==========
    name : string
        name of the object
    mar_env : instance of market_environment
        market environment data for simulation
    corr : boolean
        True if correlated with other model object

    Methods
    =======
    update :
        updates parameters
    generate_paths :
        returns Monte Carlo paths given the market environment
    '''

    def __init__(self, name, mar_env, corr=False):
        super(jump_diffusion, self).__init__(name, mar_env, corr)
        try:
            self.lamb = mar_env.get_constant('lambda')
            self.mu = mar_env.get_constant('mu')
            self.delt = mar_env.get_constant('delta')
        except:
            print('Error parsing market environment.')

    def update(self, pricing_date=None, initial_value=None,
               volatility=None, lamb=None, mu=None, delta=None,
               final_date=None):
        if pricing_date is not None:
            self.pricing_date = pricing_date
            self.time_grid = None
            self.generate_time_grid()
        if initial_value is not None:
            self.initial_value = initial_value
        if volatility is not None:
            self.volatility = volatility
        if lamb is not None:
            self.lamb = lamb
        if mu is not None:
            self.mu = mu
        if delta is not None:
            self.delt = delta
        if final_date is not None:
            self.final_date = final_date
        self.instrument_values = None

    def generate_paths(self, fixed_seed=False, day_count=365.):
        if self.time_grid is None:
            self.generate_time_grid()
            # method from generic model simulation class
        # number of dates for time grid
        M = len(self.time_grid)
        # number of paths
        I = self.paths
        # array initialization for path simulation
        paths = np.zeros((M, I))
        # initialize first date with initial_value
        paths[0] = self.initial_value
        if self.correlated is False:
            # if not correlated generate random numbers
            sn1 = sn_random_numbers((1, M, I),
                                    fixed_seed=fixed_seed)
        else:
            # if correlated use random number object as provided
            # in market environment
            sn1 = self.random_numbers

        # Standard normally distributed seudo-random numbers
        # for the jump component
        sn2 = sn_random_numbers((1, M, I),
                                fixed_seed=fixed_seed)

        forward_rates = self.discount_curve.get_forward_rates(
            self.time_grid, self.paths, dtobjects=True)[1]

        rj = self.lamb * (np.exp(self.mu + 0.5 * self.delt ** 2) - 1)
        for t in range(1, len(self.time_grid)):
                        # select the right time slice from the relevant
            # random number set
            if self.correlated is False:
                ran = sn1[t]
            else:
                # only with correlation in portfolio context
                ran = np.dot(self.cholesky_matrix, sn1[:, t, :])
                ran = ran[self.rn_set]
            dt = (self.time_grid[t] - self.time_grid[t - 1]).days / day_count
            # difference between two dates as year fraction
            poi = np.random.poisson(self.lamb * dt, I)
            # Poisson distributed pseudo-random numbers for jump component
            rt = (forward_rates[t - 1] + forward_rates[t]) / 2
            paths[t] = paths[t - 1] * (
                np.exp((rt - rj - 0.5 * self.volatility ** 2) * dt +
                       self.volatility * np.sqrt(dt) * ran) +
                (np.exp(self.mu + self.delt * sn2[t]) - 1) * poi)
        self.instrument_values = paths

jump_diffusion 类的特殊市场环境元素

元素	类型	强制	描述
lambda	常量	是	跳跃密度（概率、按年）
mu	常量	是	预期跳跃规律
delta	常量	是	跳跃规律的标准差

16.4.2 用例

me_jd = market_environment('me_jd', dt.datetime(2018, 1, 1))
me_jd.add_constant('lambda', 0.3)
me_jd.add_constant('mu', -0.75)
me_jd.add_constant('delta', 0.1)
me_jd.add_environment(me_gbm)
jd = jump_diffusion('jd', me_jd)

%time paths_3 = jd.get_instrument_values()
# Wall time: 21 ms

jd.update(lamb=0.9)
%time paths_4 = jd.get_instrument_values()
# Wall time: 19 ms

plt.figure(figsize=(8, 6))
p1 = plt.plot(gbm.time_grid, paths_3[:, :10], 'b')
p2 = plt.plot(gbm.time_grid, paths_4[:, :10], 'r-.')
plt.grid(True)
l1 = plt.legend([p1[0], p2[0]], ['low volatility', 'high volatility'], loc=2)
plt.gca().add_artist(l1)
plt.xticks(rotation=30)

来自跳跃扩散模拟类的模拟路径

16.5 平方根扩散

公式 平方根扩散的随机微分方程

dxt=k(θ−xt)dt+σxt−−√dZt d x t = k ( θ − x t ) d t + σ x t d Z t

xt x t ：日期 t 的过程水平
k ： 均值回归因子
θ θ ：长期过程均值
σ σ ：恒定波动率参数
Z ：标准布朗运动

众所周知， xt x t 的值呈卡方分布。但是，许多金融模型可以使用正态分布进行离散化和近似计算（即所谓的欧拉离散化格式）。虽然欧拉格式对几何布朗运动很准确 ，但是对于大部分其他随机过程则会产生偏差。即使有精确的格式，因为数值化或者计算的原因，使欧拉格式可能最合适。定义 s≡t−Δt s ≡ t − Δ t 和 x+≡max(x,0) x + ≡ m a x ( x , 0 ) ，下面公式提出了一种欧拉格式。这种特殊格式在文献中通常称作完全截断。
公式 平方根扩散的欧拉离散化

x~t=x~s+k(θ−x~+s)Δt+σx~+s−−−√Δtzt−−−−√ x ~ t = x ~ s + k ( θ − x ~ s + ) Δ t + σ x ~ s + Δ t z t

xt=x~+t x t = x ~ t +

16.5.1 模拟类

#
# DX Analytics
# Base Classes and Model Classes for Simulation
# square_root_diffusion.py
#

class square_root_diffusion(simulation_class):
    ''' Class to generate simulated paths based on
    the Cox-Ingersoll-Ross (1985) square-root diffusion.

    Attributes
    ==========
    name : string
        name of the object
    mar_env : instance of market_environment
        market environment data for simulation
    corr : boolean
        True if correlated with other model object

    Methods
    =======
    update :
        updates parameters
    generate_paths :
        returns Monte Carlo paths given the market environment
    '''

    def __init__(self, name, mar_env, corr=False):
        super(square_root_diffusion, self).__init__(name, mar_env, corr)
        try:
            self.kappa = mar_env.get_constant('kappa')
            self.theta = mar_env.get_constant('theta')
        except:
            print('Error parsing market environment.')

    def update(self, pricing_date=None, initial_value=None, volatility=None,
               kappa=None, theta=None, final_date=None):
        if pricing_date is not None:
            self.pricing_date = pricing_date
            self.time_grid = None
            self.generate_time_grid()
        if initial_value is not None:
            self.initial_value = initial_value
        if volatility is not None:
            self.volatility = volatility
        if kappa is not None:
            self.kappa = kappa
        if theta is not None:
            self.theta = theta
        if final_date is not None:
            self.final_date = final_date
        self.instrument_values = None

    def generate_paths(self, fixed_seed=True, day_count=365.):
        if self.time_grid is None:
            self.generate_time_grid()
        M = len(self.time_grid)
        I = self.paths
        paths = np.zeros((M, I))
        paths_ = np.zeros_like(paths)
        paths[0] = self.initial_value
        paths_[0] = self.initial_value
        if self.correlated is False:
            rand = sn_random_numbers((1, M, I),
                                     fixed_seed=fixed_seed)
        else:
            rand = self.random_numbers

        for t in range(1, len(self.time_grid)):
            dt = (self.time_grid[t] - self.time_grid[t - 1]).days / day_count
            if self.correlated is False:
                ran = rand[t]
            else:
                ran = np.dot(self.cholesky_matrix, rand[:, t, :])
                ran = ran[self.rn_set]

            # full truncation Euler discretization
            paths_[t] = (paths_[t - 1] + self.kappa *
                         (self.theta - np.maximum(0, paths_[t - 1])) * dt +
                         np.sqrt(np.maximum(0, paths_[t - 1])) *
                         self.volatility * np.sqrt(dt) * ran)
            paths[t] = np.maximum(0, paths_[t])
        self.instrument_values = paths

square_root_diffusion 类市场环境的特定元素

元素	类型	强制	描述
kappa	常量	是	均值回归因子
theta	常量	是	过程长期均值

16.5.2 用例

me_srd = market_environment('me_srd', dt.datetime(2018, 1, 1))
me_srd.add_constant('initial_value', 0.25)
me_srd.add_constant('volatility', 0.05)
me_srd.add_constant('final_date', dt.datetime(2018,12,31))
me_srd.add_constant('currency', 'EUR')
me_srd.add_constant('frequency', 'W')
me_srd.add_constant('paths', 10000)

me_srd.add_constant('kappa',4.0)
me_srd.add_constant('theta',0.2)
me_srd.add_curve('discount_curve',constant_short_rate('r',0.0))
srd=square_root_diffusion('srd',me_srd)
srd_paths=srd.get_instrument_values()[:,:10]

plt.figure(figsize=(8, 6))
p1 = plt.plot(srd.time_grid, srd.get_instrument_values()[:, :10])
plt.axhline(me_srd.get_constant('theta'), color='r', ls='--', lw=2.0)
plt.grid(True)
plt.xticks(rotation=30)

来自平方根扩融模拟类的模拟路径（虚线＝长期均值 theta）

^_^