Can We Use Mixture Models to Predict Market Bottoms?
This repo contains the code and supporting data to the webinar presented via QuantInsti.com April 25, 2017
Clone or download the repo
To import the GMM prediction dataframe and output
model_data = data/GMM_Results_TidyData.h5
import pandas as pd
df = pd.read_hdf(model_data, 'table')
print(df.head())
lookback of 999 indicates an expanding lookback window.
To import the raw return data:
market_data = data/mixture_model_merged_data_03.h5
import pandas as pd
df = pd.read_hdf(market_data, 'table')
print(df.head())
To run the model yourself and experiment with parameters:
import ModelRunner
import ResultEval
import ModelPlots
DIR = '/YOUR/PROJECT/DIR/'
# Model Params
# ------------
f1 = 'TEDRATE' # ted spread
f2 = 'T10Y2Y' # constant maturity ten yer - 2 year
f3 = 'T10Y3M' # constant maturity 10yr - 3m
factors = [f1, f2, f3]
a, b = (.2, .7) # found via coarse parameter search
alpha = 0.99
max_iter = 50
init = 'random' # or 'kmeans'
nSamples = 1_000
year = 2007 # cutoff
lookback = 1 # years
# k = 3 # n_components
# step_fwd = 3 # days
step_fwds = [1, 2, 3, 5, 7, 10, 21]
ks = [2,3,5,7,9,13,17,21]
chosen_syms = ['SPY', 'QQQ', 'TLT', 'GLD', 'EFA', 'EEM']
for k in ks:
for step in step_fwds:
for mkt in chosen_syms:
p('-'*79)
p('fitting:', mkt)
p(f'params: k = {k} | step = {step} | lookback = {lookback}')
p('...')
ft_cols = [mkt + '_lret'] + factors
MR_kwds = dict(ft_cols=ft_cols, k=k, init=init, max_iter=max_iter)
MR = ModelRunner.ModelRunner(data, **MR_kwds)
dct = MR.prediction_cycle(year, alpha, a, b,
nSamples, lookback, mkt)
res = ResultEval.ResultEval(dct, step_fwd=step)
event_dict = res.get_event_states()
event = list(event_dict.keys())[1] #[1] # TOO_LOW
post_events = res.get_post_events(event_dict[event])
end_vals = res.get_end_vals(post_events)
smry = res.create_summary(end_vals)
p()
p('*'*25)
p(mkt, event.upper())
p(smry.T)
mp = ModelPlots.ModelPlots(mkt, post_events, event, DIR, year)
agg_tmp_df = mp._agg_temp_event_returns()
mp.plot_pred_results(dct['pred'], dct['year'], dct['a'], dct['b'])
mp.plot_equity_curve(agg_tmp_df, benchmark=data['SPY_lret'])
mp.plot_distplot(end_vals, smry)
break
break
break
