def runModelSuite(N, M, P, C, alpha, nu, myRho, rhoTarget, tenor, modelType):
startTime = time.time()
if modelType == 0: # Binomial (independent-default) model
el, ul, var, es = bp.independentBinomialSimulation(N, M, P, C, alpha)
simTime = (time.time() - startTime)
elif modelType == 1: # Gaussian threshold model
el, ul, var, es = th.oneFactorThresholdModel(N, M, P, C, myRho, nu,
alpha, 0)
simTime = (time.time() - startTime)
elif modelType == 2: # Beta-binomial mixture model
myP = np.mean(P)
a, b = mix.betaCalibrate(myP, rhoTarget)
M1 = mix.betaMoment(a, b, 1)
M2 = mix.betaMoment(a, b, 2)
print("a, b parameters are %0.1f and %0.1f." % (a, b))
print("Targeted: %0.4f and calibrated: %0.4f default probability." %
(myP, M1))
print("Targeted: %0.3f and calibrated: %0.3f default correlation." %
(rhoTarget, np.divide(M2 - M1**2, M1 - M1**2)))
el, ul, var, es = mix.betaBinomialSimulation(N, M, C, a, b, alpha)
simTime = (time.time() - startTime)
elif modelType == 3: # t-distributed threshold model
el, ul, var, es = th.oneFactorThresholdModel(N, M, P, C, myRho, nu,
alpha, 1)
simTime = (time.time() - startTime)
elif modelType == 4: # Basel IRB approach
mAdjustedC = np.multiply(C, getMaturityAdjustment(tenor, P))
el = np.dot(P, mAdjustedC)
var = getBaselRiskCapital(P, tenor, C, alpha)
ul = np.sum(P * (1 - P) * mAdjustedC)
es = var
simTime = (time.time() - startTime)
elif modelType == 5: # Asymptotic Gaussian threshold model (ASRF)
meanP = np.maximum(0.0009, np.median(P))
a, b = th.getAsrfMoments(meanP, myRho)
el = np.sum(C) * a
ul = np.sum(C) * b
pdf, cdf, var, es = th.asrfModel(meanP, myRho, C, alpha)
simTime = (time.time() - startTime)
return el, ul, var, es, simTime
import varianceReduction as vr
plt.close("all")
# Model input and parameters
c = np.load(expFile)
p = np.load(dpFile)
N = len(c)
alpha = np.array([0.95, 0.97, 0.99, 0.995, 0.999, 0.9997, 0.9999])
myP = np.mean(p)
# See ex04.py for the calibration of the t-threshold model
rhoT = 0.06339367516538337
nu = 8.0269146874964417
M = 1000000
S = 10 # Caution, not very many.
print("Running t-THRESHOLD MODEL. Caution: it's a bit slow!")
el, ul, var, es = th.oneFactorThresholdModel(N, M, p, c, rhoT, nu, alpha, 1)
VaRC = vc.myVaRCYT(var[-1], p, c, rhoT, nu, 6)
print("Printing RAW MONTE-CARLO RESULTS")
C, V, E = vc.mcThresholdTDecomposition(N, M, S, p, c, rhoT, nu, 1, alpha[-1])
mcVaRC = np.mean(C, 1)[:, 0]
print("Printing IMPORTANCE-SAMPLING RESULTS")
Mis = 24000
eps = 5
tailProb = np.zeros([S])
varTarget = np.mean(V)
isContributions = np.zeros([N, S])
for s in range(0, S):
print("Running IS Estimator: %d" % (s + 1))
testIS, thetaZStar, pZ, qZ, cgf, rnDerivative = vr.isThresholdContr(
N, Mis, p, c, varTarget, rhoT, nu, -0.2)
L_is = np.dot(c, testIS)
def findAlphaGaussian(a, N, M, p, c, l, myRho):
elTemp, ulTemp, varTemp, esTemp = th.oneFactorThresholdModel(
N, M, p, c, myRho, 0, np.array([a]), 0)
return 1e4 * (l - esTemp[0])**2
importlib.reload(vc)
importlib.reload(vr)
plt.close('all')
# Model input and parameters
c = np.load(expFile)
p = np.load(dpFile)
N = len(c)
alpha = np.array([0.95,0.97,0.99,0.995,0.999,0.9997,0.9999])
myP =np.mean(p)
# See ex04.py for the calibration of the t-threshold model
rhoT = 0.06339367516538337
nu = 8.0269146874964417
M = 1000000
S = 10 # Caution, not very many.
print("Running t-THRESHOLD MODEL. Caution: it's a bit slow!")
el,ul,var,es = th.oneFactorThresholdModel(N,M,p,c,rhoT,nu,alpha,1)
VaRC = vc.myVaRCYT(var[-1],p,c,rhoT,nu,6)
print("Printing RAW MONTE-CARLO RESULTS")
C,V,E = vc.mcThresholdTDecomposition(N,M,S,p,c,rhoT,nu,1,alpha[-1])
mcVaRC = np.mean(C,1)[:,0]
print("Printing IMPORTANCE-SAMPLING RESULTS")
Mis = 24000
eps = 5
tailProb = np.zeros([S])
varTarget = np.mean(V)
isContributions = np.zeros([N,S])
for s in range(0,S):
print("Running IS Estimator: %d" % (s+1))
testIS,thetaZStar,pZ,qZ,cgf,rnDerivative = \
vr.isThresholdContr(N,Mis,p,c,varTarget,rhoT,nu,-0.2)
L_is = np.dot(c,testIS)
rhoG = resultG.x
# (b) Simulate
startTime = time.time()
el[0], ul[0], var[:, 0], es[:, 0] = th.oneFactorGaussianModel(
N, M, p, c, rhoG, alpha)
cTime[0] = (time.time() - startTime)
print("Running t THRESHOLD MODEL")
# (a) Calibrate
tModel = scipy.optimize.fsolve(th.tCalibrate,
np.array([startRho, startNu]),
args=(myP, rhoTarget, tDependenceTarget))
rhoT = tModel[0]
nu = tModel[1]
# (b) Simulate
startTime = time.time()
el[1], ul[1], var[:, 1], es[:, 1] = th.oneFactorThresholdModel(
N, M, p, c, rhoT, nu, alpha, 1)
cTime[1] = (time.time() - startTime)
print("Running VARIANCE-GAMMA MODEL")
# (a) Calibrate
kTarget = 6
vgModel = scipy.optimize.fsolve(th.nvmCalibrate,
np.array([0.2, 1]),
args=(myP, rhoTarget, kTarget, 0))
rhoVG = vgModel[0]
vgA = vgModel[1]
nvmOutcome = th.nvmCalibrate(np.array([rhoVG, vgA]), myP, 0, 0, 0)
# (b) Simulate
startTime = time.time()
el[2], ul[2], var[:, 2], es[:,
2] = th.oneFactorNVMModel(N, M, p, c, rhoVG, vgA,
alpha, 0)