monte_runs = 1000
for want_rho in np.arange(-0.8, 0.8, 0.1):
for want_sharpe in [0.0, 0.5, 1.0, 2.0]:
for want_stdev in [1.0]:
for want_skew in [-2.0, -1.0, 0.0, 0.5, 1.0]:
want_mean = want_stdev * want_sharpe / 16.0
## autocorrelation introduces biases
(adj_want_mean, adj_want_skew,
adj_want_stdev) = adj_moments_for_rho(want_rho, want_mean,
want_skew, want_stdev)
manyans = [
autocorr_skewed_returns(want_rho, adj_want_mean,
adj_want_stdev, adj_want_skew)
for notused in range(monte_runs)
]
sample_mean = np.mean([np.mean(x) for x in manyans])
sample_std = np.mean([np.std(x) for x in manyans])
sample_skew = np.mean([st.skew(x) for x in manyans])
sample_rho = np.mean([autocorr(x) for x in manyans])
print "****************************"
print "Mean, wanted %.2f got %.6f" % (want_mean, sample_mean)
print "Stdev, wanted %.2f got %.4f" % (want_stdev, sample_std)
print "Skew, wanted %.2f got %.4f" % (want_skew, sample_skew)
print "Rho, wanted %.2f got %.4f" % (want_rho, sample_rho)
want_skew=0.0
want_Sharpes=[0.5, 1.0]
length_period_years=20
# calculations
length_period=length_period_years*DAYS_IN_YEAR
want_ann_stdev=0.2
want_ann_mean=[ann_SR*want_ann_stdev for ann_SR in want_Sharpes]
want_mean_list=[this_ann_mean/DAYS_IN_YEAR for this_ann_mean in want_ann_mean]
want_stdev=want_ann_stdev/ROOT_DAYS_IN_YEAR
happywithcurve=False
while not happywithcurve:
## generate a new Curve
equitycurves=[arbitrary_timeseries(autocorr_skewed_returns(rho, want_mean, want_stdev,
want_skew, size=length_period), index_start=pd.datetime(1995,1,1))
for want_mean in want_mean_list]
equitycurves_pd=pd.concat(equitycurves, axis=1)
equitycurves_pd.columns=["A", "B"]
equitycurves_pd.cumsum().plot()
show()
ans=raw_input("Happy with this? Y / N? ")
if ans=="Y":
happywithcurve=True
sliced_data=slices_for_ts(equitycurves[0])
start_point=sliced_data[0]
for idx in range(len(sliced_data))[1:]:
sliced_curves=[eq_curve[start_point:sliced_data[idx]] for eq_curve in equitycurves]
equitycurves_pd=pd.concat(sliced_curves, axis=1)
avg_annual_returns_this_scenario = []
avg_drawdowns_this_scenario = []
max_drawdowns_this_scenario = []
avg_avg_ratio_this_scenario = []
avg_max_ratio_this_scenario = []
## Window '1000' is no ovelay at all
for N_length in N_windows:
print "N_length: %d" % N_length
## Generate a lot of random data
many_curves_raw = [
arbitrary_timeseries(
autocorr_skewed_returns(want_rho,
adj_want_mean,
adj_want_stdev,
adj_want_skew,
size=curve_length))
for notused in range(monte_runs)
]
## Apply the overlay
many_curves = [
account_curve(apply_overlay(x, N_length, period_stdev, costs_SR))
for x in many_curves_raw
]
## Calc statistics
annual_returns_these_curves = [
np.mean(x) * period_adj for x in many_curves
]
want_skew=0.0
want_Sharpes=[0.5, 1.0]
length_period_years=20
# calculations
length_period=length_period_years*DAYS_IN_YEAR
want_ann_stdev=0.2
want_ann_mean=[ann_SR*want_ann_stdev for ann_SR in want_Sharpes]
want_mean_list=[this_ann_mean/DAYS_IN_YEAR for this_ann_mean in want_ann_mean]
want_stdev=want_ann_stdev/ROOT_DAYS_IN_YEAR
happywithcurve=False
while not happywithcurve:
## generate a new Curve
equitycurves=[arbitrary_timeseries(autocorr_skewed_returns(rho, want_mean, want_stdev,
want_skew, size=length_period), index_start=pd.datetime(1995,1,1))
for want_mean in want_mean_list]
equitycurves_pd=pd.concat(equitycurves, axis=1)
equitycurves_pd.columns=["A", "B"]
equitycurves_pd.cumsum().plot()
show()
ans=raw_input("Happy with this? Y / N? ")
if ans=="Y":
happywithcurve=True
sliced_data=slices_for_ts(equitycurves[0])
start_point=sliced_data[0]
for idx in range(len(sliced_data))[1:]:
sliced_curves=[eq_curve[start_point:sliced_data[idx]] for eq_curve in equitycurves]
equitycurves_pd=pd.concat(sliced_curves, axis=1)
"""This code verifies that the single frequency method for generating
returns with a given distribution skew, autocorrelation; all works...
"""
monte_runs=1000
for want_rho in np.arange(-0.8, 0.8, 0.1):
for want_sharpe in [0.0, 0.5, 1.0, 2.0]:
for want_stdev in [1.0]:
for want_skew in [-2.0, -1.0, 0.0, 0.5, 1.0]:
want_mean=want_stdev*want_sharpe/16.0
## autocorrelation introduces biases
(adj_want_mean, adj_want_skew, adj_want_stdev)=adj_moments_for_rho(want_rho, want_mean, want_skew, want_stdev)
manyans=[autocorr_skewed_returns(want_rho, adj_want_mean, adj_want_stdev, adj_want_skew) for notused in range(monte_runs)]
sample_mean=np.mean([np.mean(x) for x in manyans])
sample_std=np.mean([np.std(x) for x in manyans])
sample_skew=np.mean([st.skew(x) for x in manyans])
sample_rho=np.mean([autocorr(x) for x in manyans])
print "****************************"
print "Mean, wanted %.2f got %.6f" % (want_mean, sample_mean)
print "Stdev, wanted %.2f got %.4f" % (want_stdev, sample_std)
print "Skew, wanted %.2f got %.4f" % (want_skew, sample_skew)
print "Rho, wanted %.2f got %.4f" % (want_rho, sample_rho)
print "Scenario:"
print scenario
(notUsed, want_mean, want_stdev, want_skew, want_rho)=scenario
(adj_want_mean, adj_want_skew, adj_want_stdev)=adj_moments_for_rho(want_rho, want_mean, want_skew, want_stdev)
avg_annual_returns_this_scenario=[]
avg_drawdowns_this_scenario=[]
max_drawdowns_this_scenario=[]
avg_avg_ratio_this_scenario=[]
avg_max_ratio_this_scenario=[]
## Window '1000' is no ovelay at all
for N_length in N_windows:
print "N_length: %d" % N_length
## Generate a lot of random data
many_curves_raw=[arbitrary_timeseries(autocorr_skewed_returns(want_rho, adj_want_mean, adj_want_stdev, adj_want_skew, size=curve_length)) for notused in range(monte_runs)]
## Apply the overlay
many_curves=[account_curve(apply_overlay(x, N_length, period_stdev, costs_SR)) for x in many_curves_raw]
## Calc statistics
annual_returns_these_curves=[np.mean(x)*period_adj for x in many_curves]
avg_drawdowns_these_curves=[x.avg_drawdown() for x in many_curves]
max_drawdowns_these_curves=[x.worst_drawdown() for x in many_curves]
## Add results
avg_annual_returns_this_scenario.append(np.mean(annual_returns_these_curves))
avg_drawdowns_this_scenario.append(np.nanmean(avg_drawdowns_these_curves))
max_drawdowns_this_scenario.append(np.nanmean(max_drawdowns_these_curves))
avg_avg_ratio_this_scenario.append(np.nanmean([divxx(x) for x in zip(annual_returns_these_curves, avg_drawdowns_these_curves)]))
avg_max_ratio_this_scenario.append(np.nanmean([divxx(x) for x in zip(annual_returns_these_curves, max_drawdowns_these_curves)]))
(notUsed, want_mean, want_stdev, want_skew, want_rho) = scenario
(adj_want_mean, adj_want_skew, adj_want_stdev) = adj_moments_for_rho(want_rho, want_mean, want_skew, want_stdev)
avg_annual_returns_this_scenario = []
avg_drawdowns_this_scenario = []
max_drawdowns_this_scenario = []
avg_avg_ratio_this_scenario = []
avg_max_ratio_this_scenario = []
## Window '1000' is no ovelay at all
for N_length in N_windows:
print "N_length: %d" % N_length
## Generate a lot of random data
many_curves_raw = [
arbitrary_timeseries(
autocorr_skewed_returns(want_rho, adj_want_mean, adj_want_stdev, adj_want_skew, size=curve_length)
)
for notused in range(monte_runs)
]
## Apply the overlay
many_curves = [account_curve(apply_overlay(x, N_length, period_stdev, costs_SR)) for x in many_curves_raw]
## Calc statistics
annual_returns_these_curves = [np.mean(x) * period_adj for x in many_curves]
avg_drawdowns_these_curves = [x.avg_drawdown() for x in many_curves]
max_drawdowns_these_curves = [x.worst_drawdown() for x in many_curves]
## Add results
avg_annual_returns_this_scenario.append(np.mean(annual_returns_these_curves))
avg_drawdowns_this_scenario.append(np.nanmean(avg_drawdowns_these_curves))
length_period = length_period_years * DAYS_IN_YEAR
want_ann_stdev = 0.2
want_ann_mean = [ann_SR * want_ann_stdev for ann_SR in want_Sharpes]
want_mean_list = [
this_ann_mean / DAYS_IN_YEAR for this_ann_mean in want_ann_mean
]
want_stdev = want_ann_stdev / ROOT_DAYS_IN_YEAR
happywithcurve = False
while not happywithcurve:
## generate a new Curve
equitycurves = [
arbitrary_timeseries(autocorr_skewed_returns(rho,
want_mean,
want_stdev,
want_skew,
size=length_period),
index_start=pd.datetime(1995, 1, 1))
for want_mean in want_mean_list
]
equitycurves_pd = pd.concat(equitycurves, axis=1)
equitycurves_pd.columns = ["A", "B"]
equitycurves_pd.cumsum().plot()
show()
ans = raw_input("Happy with this? Y / N? ")
if ans == "Y":
happywithcurve = True
sliced_data = slices_for_ts(equitycurves[0])
start_point = sliced_data[0]
