clustering forecasts

clustering_forecasts.py

import matplotlib

matplotlib.use("TkAgg")
from syscore.constants import arg_not_supplied


from systems.basesystem import System
from systems.provided.rob_system.run_system import futures_system
from sysquant.estimators.clustering_correlations import *
import pandas as pd

system = futures_system()
instrument_list = system.get_instrument_list()

all_pandl = dict()
instrument_weights = system.config.instrument_weights
for rule_name in list(system.rules.trading_rules().keys()):
    pandl_this_rule = dict()
    for instrument_code in list(instrument_weights.keys()):

        pandl_this_code_rule = system.accounts.pandl_for_instrument_forecast(
            instrument_code, rule_name
        ).percent.gross
        pandl_this_code_rule = (
            pandl_this_code_rule * instrument_weights[instrument_code]
        )
        pandl_this_rule[instrument_code] = pandl_this_code_rule

    pandl_this_rule = pd.concat(pandl_this_rule, axis=1)
    pandl_this_rule = pandl_this_rule.sum(axis=1)

    all_pandl[rule_name] = pandl_this_rule

all_pandl = pd.concat(all_pandl, axis=1)
import pickle

f = open("/home/rob/temp.pck", "wb")
pickle.dump(all_pandl, f)
f.close()

from sysquant.estimators.correlation_estimator import correlationEstimate

corr_matrix = correlationEstimate.from_pd(all_pandl.corr())


def display_clusters(system, clusters):
    for idx, cluster in enumerate(clusters):
        print("Cluster %d" % (idx + 1))

        print(cluster)


def show_cluster(system, cluster):
    instruments = cluster
    assets = [system.data.asset_class_for_instrument(code) for code in instruments]
    classes = list(set(assets))
    for aclass in classes:
        subset = [
            code
            for (code, asset_class) in zip(instruments, assets)
            if asset_class == aclass
        ]
        print("%s: %s" % (aclass, str(subset)))


N = 12
clusters = cluster_correlation_matrix(corr_matrix, N)
display_clusters(system, clusters)

## get turnovers
## doesn't matter which instrument as long as long history
turnovers = dict()
for rule_name in corr_matrix.columns:
    turnovers[rule_name] = system.accounts.forecast_turnover("SP500_micro", rule_name)

print({k: v for k, v in sorted(turnovers.items(), key=lambda item: item[1])})