''' monte carlo tool for simple strategy, more generally for use with any strategy that I want to back test... Engine takes a class instance, derived from a base class wih two methods initialise() onsimulation() aftersimulation() ontrial() finalise() ''' import unittest, datetime import numpy as np from BackTest import MonteCarloModel, MonteCarloEngine, Simulation import matplotlib.pyplot as plt from random import randint from scipy.optimize import fmin_powell from hashlib import md5 from time import localtime def Root2(r,verbose=True): ''' simple wrapper routine for solver. returns the energy level for a given R=r. the solver can use this method to minimise the energy by varying r as necessary. ''' if r <= 0: return 1e99 else: e = MonteCarloEngine(moduleName='MonteCarloHeadsTailsExample', className='SimpleHeadTailModel') losers = e.start(args={'wager_multiplier':r}) if verbose: print 'solving for r: ', r return losers class SimpleHeadTailModel(MonteCarloModel): ''' model wager previous bet success or not, impact to profit. ''' def toss(self): ''' 1 - win -1 - loss Each toss determines whether the position is successful or not. This way no need to keep track of a decision and an associated variable. Simply do I win or not. ''' coin_toss = randint(1,2) if coin_toss == 1: return 1 else: return -1 def initialise(self, context): self.name = 'My Simple Heads And Tails Model' # start with 10 USD bet self.wager= 100 self.wager_initial= 100 self.starting_pot = 1000 self.previous_value = 1 # default to 1 on first round self.simulations = 100 # MC simulation trials self.trials = 100 # subintervals self.r = np.zeros(shape=(self.simulations, self.trials), dtype=float) # matrix to hold all results self.pnl = np.zeros(shape=(self.simulations, self.trials), dtype=float) # matrix to hold all results print "simulations %d, trials %d starting pot %d " % (self.simulations, self.trials, self.starting_pot) # Tell the engine where to associate the data to security. context[self.name] = Simulation(self.simulations, self.trials, self.toss) self.fig = plt.figure() self.ax = self.fig.add_subplot(211) self.ax1 = self.fig.add_subplot(212) self.ax.autoscale_view(True,True,True) def onsimulation(self, model, simulation, engine): self.r[simulation,0] = 0 # assume starting pot here self.pnl[simulation,0] = self.starting_pot def aftersimulation(self, model, simulation, engine): self.ax.plot(np.arange(0, self.trials, 1), self.r[simulation]) self.ax1.plot(np.arange(0, self.trials, 1), self.pnl[simulation]) def reset_wager(self): self.wager = self.wager_initial def ontrial(self, model, simulation, trial, value, engine, args): ''' want to test some strategies for betting set wager for each bet if previous bet value : float sample from model ''' # if we lost last time then double up if self.previous_value == -1: if args.has_key('args'): self.wager = (self.wager*float(args['args']['wager_multiplier'][0])) else: self.wager = (self.wager*0.1) # keep track of coin toss paths self.r[simulation,trial] = self.r[simulation,trial-1] + value if args.has_key('args'): self.r0 = float(args['args']['wager_multiplier'][0]) else: self.r0 =0.1 # if we won, add the wager # else subtract the wager if self.pnl[simulation,trial-1] > 0: if value == 1 : self.pnl[simulation,trial] = self.pnl[simulation,trial-1] + self.wager else: self.pnl[simulation,trial] = self.pnl[simulation,trial-1] - self.wager else: # no bet to be made here self.pnl[simulation,trial] = self.pnl[simulation,trial-1] # always reset wager self.reset_wager() # keep track of the previous value for next time around self.previous_value = value def add_prefix(self, filename): from hashlib import md5 from time import localtime return "%s_%s"%(md5(str(localtime())).hexdigest(), filename) def finalise(self, model, engine): ''' returns the value that we are trying to minimise, here the number of losers. ''' # what is our survivability number_of_losers = len([f for f in self.pnl if f[len(f)-1]<=0]) number_of_survivors = len([f for f in self.pnl if f[len(f)-1]>0]) number_of_participants = len(self.pnl) print "participants [%d] survivors [%2.1f%%] losers [%2.1f%%] weight [%2.6f] "% (number_of_participants, float(number_of_survivors)/float(number_of_participants)*100, float(number_of_losers)/float(number_of_participants)*100, self.r0) plt.title('Simulations %d Steps %d' % (int(self.simulations), int(self.trials))) plt.xlabel('steps') plt.ylabel('profit and loss') plt.savefig("%s_%s"%(md5(str(localtime())).hexdigest(), 'model')) return float(number_of_losers)/float(number_of_participants)*100 class TestNode(unittest.TestCase): def setUp(self): pass def test_engine(self): ''' example of how to launch the MontoCarloTestEngine this is modelled on the quantopian style interface. ''' e = MonteCarloEngine(moduleName='MonteCarloHeadsTailsExample', className='SimpleHeadTailModel') e.start() def test_minimise(self): print '#################################' print '# Test Equilibrium Loss Wager' print '#################################' wager_multiplier=fmin_powell(Root2, x0=1., maxiter=20) print "highest survivability following loss, multiply wager by %2.4f %% "%(wager_multiplier*100) if __name__ == '__main__': unittest.main() ================================================================== == BackTest module ================================================================== ''' back testing tool for prediction strategy, more generally for use with any strategy that I want to back test... Engine takes a class instance, derived from a base class wih twomethods initialise() ondata() ''' import unittest, time, datetime from pandas import DataFrame import numpy as np from pylab import show import random from abc import ABCMeta, abstractmethod import pandas as pd # # Simple python component wrapper # class Component(object): __metaclass__ = ABCMeta @abstractmethod def start(self): raise NotImplementedError("Should implement intialise()!") class BackTestModel(object): __metaclass__ = ABCMeta @abstractmethod def initialise(self, context): raise NotImplementedError("Should implement intialise()!") @abstractmethod def ondata(self, sid, data): raise NotImplementedError("Should implement ondata()!") class MonteCarloModel(object): __metaclass__ = ABCMeta @abstractmethod def initialise(self, context): raise NotImplementedError("Should implement intialise()!") @abstractmethod def ontrial(self, model, simulation, trial, value, engine, args): raise NotImplementedError("Should implement ontrial()!") def onsimulation(self, model, simulation, engine): raise NotImplementedError("Should implement onsimulation()!") def aftersimulation(self, model, simulation, engine): raise NotImplementedError("Should implement aftersimulation()!") def finalise(self, model, engine): raise NotImplementedError("Should implement finalise()!") class Simulation(object): ''' simple wrapper that describes a simulation ''' def __init__(self, n, m, func): ''' n integer number of simulations m integer number of trials per simulation func class method or function used to sample the value ''' self.number_of_simulations = n self.number_of_trials = m self.func = func @property def sample(self): return self.func class MonteCarloEngineException(Exception): pass class MonteCarloEngine(Component): ''' twist on the Engine that will take a different type of context, this time a Simulation class instance. This will be called class ExampleModel(MonteCarloModel): def initialise(self, context): context['My Simple Model'] = Simulation(10, 100, standard_normal()) print 'setting My Simple Model' def ondata(self, model, simulation, trial, value, engine): print simulation, trial, value, engine ''' def __init__(self, moduleName, className): self.context = {} self.obj = self.__generate__(moduleName, className) #print isinstance(self.obj, MonteCarloModel), hasattr(self.obj, 'initialise') if isinstance(self.obj, MonteCarloModel): if hasattr(self.obj, 'initialise'): self.obj.initialise(self.context) print 'calling initialise' else: print 'no initialise' # # TODO: load data from somewhere for securities in context # print self.context def __generate__(self, module_name, class_name): module = __import__(module_name) class_ = getattr(module, class_name) instance = class_() #print instance return instance def start(self, **args): #print 'starting...', self.obj, self.context, args if self.obj is None: raise MonteCarloEngineException('No engine exists') for name, model in self.context.items(): for simulation in np.arange(0, model.number_of_simulations): # number of MC simulations # call to signal new simulation if hasattr(self.obj, 'onsimulation'): self.obj.onsimulation(model, simulation, self) for trial in np.arange(1,model.number_of_trials): #trials per simulation value = model.sample() if hasattr(self.obj, 'ontrial'): self.obj.ontrial(model, simulation, trial, value, self, args) # call to signal after simulation if hasattr(self.obj, 'aftersimulation'): self.obj.aftersimulation(model, simulation, self) #self.post_ondata(k, index2, value) if hasattr(self.obj, 'finalise'): return self.obj.finalise(model, self) ''' class ExampleModel(BackTestModel): def initialise(self, context): context['ARM.L'] = 'Book1.csv' print 'setting ARM.L' def ondata(self, sid, data): print sid, data ''' class Engine(Component): ''' responsible for handling instances of the back test models ''' def __init__(self, moduleName, className): self.context = {} self.data = {} self.orders = {} self.positions = {} self.pnl = {} self.risk = {} self.obj = self.__generate__(moduleName, className) #print isinstance(self.obj, BackTestModel), hasattr(self.obj, 'initialise') if isinstance(self.obj, BackTestModel): if hasattr(self.obj, 'initialise'): self.obj.initialise(self.context) #print 'calling initialise' else: print 'no initialise' # # TODO: load data from somewhere for securities in context # #print self.context for k in self.context.keys(): #print k t = time.clock() myfilename = self.context[k] data = DataFrame.from_csv(myfilename,header=0,index_col=0,parse_dates=True) print 'load data', time.clock()-t self.data[k] = data self.positions[k] = 0 self.pnl[k] = 0 self.risk[k] = 0 def order(self, sid, value): # queue order to be processed self.orders[sid] = (value, False) @property def position(self): return self.positions def __generate__(self, module_name, class_name): module = __import__(module_name) class_ = getattr(module, class_name) instance = class_() print instance return instance def post_ondata(self, sid, index, value): # process orders for k in self.context.keys(): if hasattr(self, 'orders'): if len(self.orders) == 0: print 'no orders' break m_size, m_processed = self.orders[k] # check we have not processed this order already. if not m_processed: if not (self.positions[k]+ m_size <= 0): self.positions[k] += m_size print 'ordering %d of %s, total %d' % (m_size, k, self.positions[k]) self.orders[k] = None else: print 'no short selling' # handle pnl and risk # tick() charts def start(self): print 'starting...', self.obj, self.context if not self.obj is None: for k in self.context: for i, (index, value) in enumerate(self.data[k]['value'].iteritems()): index2 = datetime.datetime(pd.to_datetime(index).year , pd.to_datetime(index).month , pd.to_datetime(index).day) self.obj.ondata(k , index2 , value , self.data[k]['value'][0:i] , self) self.post_ondata(k, index2, value)