def __init__(self, params, randomstate=None):
self.__params = params
self.__randomstate = npu.getrandomstate() if randomstate is None else randomstate
self.__oneminuspersistence = 1. - params.persistence
self.__jumpvar = params.jumpvol * params.jumpvol
self.__oneminusjumpintensity = 1. - params.jumpintensity
def __init__(self, params, randomstate=None):
self.__scale = params.voloflogvar / np.sqrt(1. - params.persistence *
params.persistence)
randomstate = npu.getrandomstate(
) if randomstate is None else randomstate
self.__normalvariatesgenerator = rnd.NormalVariatesGenerator(
randomstate)
def __init__(self, params, randomstate=None):
self.__params = params
self.__randomstate = npu.getrandomstate() if randomstate is None else randomstate
self.__oneminuspersistence = (1. - self.__params.persistence)
self.__meanlogvartimesoneminuspersistence = self.__params.meanlogvar * self.__oneminuspersistence
self.__vc = self.__params.voloflogvar * self.__params.cor
self.__scalar = self.__params.voloflogvar * np.sqrt(1. - self.__params.cor * self.__params.cor)
def __init__(self, mean, covariance, randomstate=None):
self.__mean = npu.immutablecopyof(npu.tondim1(mean))
self.__covariance = npu.immutablecopyof(
npp.checkshapeissquare(npu.tondim2(covariance)))
self.__randomstate = npu.getrandomstate(
) if randomstate is None else randomstate
self.__impl = stats.multivariate_normal(self.__mean, self.__covariance)
def __init__(self, timecount, ar, ma, var=1., const=0., randomstate=None):
self.__timecount = timecount
self.__ar = npu.tondim1(npu.immutablecopyof(ar))
self.__ma = npu.tondim1(npu.immutablecopyof(ma))
self.__var = var
self.__const = const
self.__randomstate = npu.getrandomstate() if randomstate is None else randomstate
def __init__(self,
timecount,
params,
cortiming,
logreturnforward,
logreturnscale,
randomstate=None,
usestratonovichcorrection=False):
self.__timecount = timecount
self.__params = params
self.__cortiming = cortiming
self.__logreturnforward = logreturnforward
self.__logreturnscale = logreturnscale
self.__randomstate = npu.getrandomstate(
) if randomstate is None else randomstate
self.__logvarinitialdistribution = LogVarInitialDistribution(
params, self.__randomstate)
self.__usestratonovichcorrection = usestratonovichcorrection
kde = bc.kde(nodenames)
print('KDE bandwidths for these parameters:', kde.bw)
paramposterior = kde.pdf(
[params.meanlogvar, params.persistence, params.cor, params.voloflogvar])
logparamposterior = np.log(paramposterior)
print('Loading SV data...')
svdata = sv.loading.loadSVDataFromBUGSDataset(observationsfile,
logreturnforward=True,
logreturnscale=100.)
print(svdata)
fig = plt.figure()
sv.visualisation.makesvdataplot(fig, svdata)
randomstate = npu.getrandomstate()
if model == 'svl':
filterrundata = runsvljparticlefilter(svdata, params, randomstate)
elif model == 'svl2':
filterrundata = runsvl2particlefilter(svdata, params, randomstate)
print(filterrundata)
fig = plt.figure()
filtering.visualisation.makeparticlehistogram(fig, filterrundata.stochfilter)
fig = plt.figure()
filtering.visualisation.makefilterrunplot(fig, filterrundata.filterrundf)
loglikelihood = filterrundata.summary()['log-likelihood']
print('log-likelihood:', loglikelihood)
print('log of params posterior:', logparamposterior)
def __init__(self, params, randomstate=None):
self.__params = params
self.__randomstate = npu.getrandomstate() if randomstate is None else randomstate
def __init__(self, params, randomstate=None):
self.__params = params
self.__randomstate = npu.getrandomstate() if randomstate is None else randomstate
self.__meanlogvartimesoneminuspersistence = self.__params.meanlogvar * (1. - self.__params.persistence)
def main():
np.seterr(divide='raise', invalid='raise')
randomstate = npu.getrandomstate()
params = sv.Params(
meanlogvar=.65762,
persistence=.96125,
voloflogvar=np.sqrt(0.020053), # 0.1416
cor=-.19,
#jumpintensity=0.01,
jumpintensity=0.,
jumpvol=0.01)
#params = sv.Params(meanlogvar=-0.2076, persistence=0.9745, cor=0.0, voloflogvar=0.0492, jumpintensity=0., jumpvol=1.)
# SVL, ds1 -924.077823959 (600 particles)
# params = sv.Params(meanlogvar=-0.5486, persistence=0.9861, cor=-0.1969, voloflogvar=0.149, jumpintensity=0., jumpvol=1.)
# SVL, ds2 -2721.36910265 (600 particles)
# params = sv.Params(meanlogvar=-0.1706, persistence=0.9755, cor=-0.2699, voloflogvar=0.1464, jumpintensity=0., jumpvol=1.)
# SVL2, ds1 -923.050833581 (1000 particles)
# params = sv.Params(meanlogvar=-0.5883, persistence=0.9853, cor=-0.1472, voloflogvar=0.1456, jumpintensity=0., jumpvol=1.)
# SVL2, ds2 -2723.29157267 (1000 particles)
# params = sv.Params(meanlogvar=-0.2076, persistence=0.9745, cor=-0.275, voloflogvar=0.1492 * 1.25, jumpintensity=0., jumpvol=1.)
"""
params = sv.Params(
meanlogvar = 0.25,
persistence = 0.975,
#cor = -0.8,
cor = -0.5,
voloflogvar = np.sqrt(0.025),
jumpintensity = 0.0,
jumpvol = 10.)
"""
"""
params = sv.Params(
meanlogvar = 0.2048,
persistence = 0.6726,
#cor = -0.8,
cor = 0.004101,
voloflogvar = 17.62,
jumpintensity = 0.0,
jumpvol = 10.)
"""
params = sv.Params(
meanlogvar=-3.971,
persistence=0.2338,
#cor = -0.8,
cor=-0.9178,
voloflogvar=0.01468,
jumpintensity=0.0,
jumpvol=10.)
# Wrong!
# params = sv.Params(meanlogvar=-5.2076, persistence=0.9745, cor=0.275, voloflogvar=1.1492, jumpintensity=0., jumpvol=1.)
# params = sv.Params(meanlogvar=-0.2076, persistence=0.9745, cor=-0.275, voloflogvar=0.1492, jumpintensity=0.01, jumpvol=10.)
initialprice = 100.
timecount = 2001 # 0, 1, ..., 1000
print('Generating SV data...')
svdata = generatesvdata(params, timecount, randomstate)
print('Loading SV data...')
# filepath = r"C:\Users\Paul\Documents\dev\alexandria\bilokon-msc\dissertation\code\winbugs\datasets\dataset-1_GBPUSD_1981-10-01_1985-06-28.txt"
# filepath = r"C:\Users\Paul\Documents\dev\alexandria\bilokon-msc\dissertation\code\datasets\dataset-2_y.txt"
filepath = r"C:\Users\Paul\Documents\dev\alexandria\bilokon-msc\dissertation\code\datasets\dataset-14-ESM16\dataset-14-ESM16_y.txt"
dtfilepath = r"C:\Users\Paul\Documents\dev\alexandria\bilokon-msc\dissertation\code\datasets\dataset-14-ESM16\dataset-14-ESM16_dt.txt"
svdata = sv.loading.loadSVDataFromBUGSDataset(filepath,
logreturnforward=True,
logreturnscale=100.,
dtfilepath=dtfilepath)
enrichsvdata(svdata, initialprice)
# action = 'examinesvdata'
# action = 'analyseparamsneighbourhood'
# action = 'optimiseparams'
action = 'runsvljparticlefilteronceandanalyse'
# action = 'runsvl2particlefilteronceandanalyse'
# action = 'runwcsvlparticlefilteronceandanalyse'
# action = 'runsvlgaussianfilteronceandanalyse'
# action = 'runsvl2gaussianfilteronceandanalyse'
# action = 'rununscentedkalmanfilterandanalyse'
# action = 'runkalmanfilterandanalyse'
print('Analysing SV data...')
examinesvdata(svdata)
print('Running action: %s...' % action)
if action == 'examinesvdata':
pass
elif action == 'analyseparamsneighbourhood':
analyseparamsneighbourhood(svdata,
params,
includejumps=False,
randomstate=randomstate)
elif action == 'optimiseparams':
initialguessparams = sv.Params(meanlogvar=-.1,
persistence=.975,
voloflogvar=np.sqrt(.02),
cor=-.8,
jumpintensity=0.,
jumpvol=1.)
optimiseparams(svdata,
initialguessparams,
params,
runsvljparticlefilter,
includejumps=False,
randomstate=randomstate)
else:
if action == 'runsvljparticlefilteronceandanalyse':
filterrundata = runsvljparticlefilter(svdata, params, randomstate)
fig = plt.figure()
filtering.visualisation.makeparticlehistogram(
fig, filterrundata.stochfilter)
elif action == 'runsvl2particlefilteronceandanalyse':
filterrundata = runsvl2particlefilter(svdata, params, randomstate)
fig = plt.figure()
filtering.visualisation.makeparticlehistogram(
fig, filterrundata.stochfilter)
elif action == 'runwcsvlparticlefilteronceandanalyse':
filterrundata = runwcsvlparticlefilter(svdata, params, randomstate)
fig = plt.figure()
filtering.visualisation.makeparticlehistogram(
fig, filterrundata.stochfilter)
elif action == 'runsvlgaussianfilteronceandanalyse':
filterrundata = runsvlgaussianfilter(svdata, params)
elif action == 'runsvl2gaussianfilteronceandanalyse':
filterrundata = runsvl2gaussianfilter(svdata, params)
elif action == 'rununscentedkalmanfilterandanalyse':
filterrundata = rununscentedkalmanfilter(svdata, params)
elif action == 'runkalmanfilterandanalyse':
filterrundata = runkalmanfilter(svdata, params)
else:
raise RuntimeError('Invalid action')
print(filterrundata)
fig = plt.figure()
filtering.visualisation.makefilterrunplot(fig,
filterrundata.filterrundf)
plt.show()
def __init__(self, elements, randomstate=None):
self.__elements = [e for e in elements]
self.__randomstate = npu.getrandomstate(
) if randomstate is None else randomstate
