def backtest(agent, env, market):
global PATH_prefix
print("starting to backtest......")
from agents.UCRP import UCRP
from agents.Winner import WINNER
from agents.Losser import LOSSER
agents = []
agents.extend(agent)
agents.append(WINNER())
agents.append(UCRP())
agents.append(LOSSER())
labels = ['PG', 'Winner', 'UCRP', 'Losser']
wealths_result = []
rs_result = []
for i, agent in enumerate(agents):
stocktrader = StockTrader()
info = env.step(None, None, 'False')
r, contin, s, w1, p, risk = parse_info(info)
contin = 1
wealth = 10000
wealths = [wealth]
rs = [1]
while contin:
w2 = agent.predict(s, w1)
env_info = env.step(w1, w2, 'False')
r, contin, s_next, w1, p, risk = parse_info(env_info)
wealth = wealth * math.exp(r)
rs.append(math.exp(r) - 1)
wealths.append(wealth)
s = s_next
stocktrader.update_summary(0, r, 0, 0, w2, p)
stocktrader.write(map(lambda x: str(x), env.get_codes()), labels[i])
print('finish one agent')
wealths_result.append(wealths)
rs_result.append(rs)
# print('资产名称',' ','平均日收益率',' ','夏普率',' ','最大回撤')
print('Asset Name', ' ', 'Average daily rate of return', ' ',
'Sharp rate', ' ', 'Maximum withdrawal')
plt.figure(figsize=(8, 6), dpi=100)
for i in range(len(agents)):
plt.plot(wealths_result[i], label=labels[i])
plt.title('RL - PM in the {} stock market'.format(market))
mrr = float(np.mean(rs_result[i]) * 100)
sharpe = float(
np.mean(rs_result[i]) / np.std(rs_result[i]) * np.sqrt(252))
maxdrawdown = float(
max(1 - min(wealths_result[i]) /
np.maximum.accumulate(wealths_result[i])))
print(labels[i], ' ', round(mrr, 3), '%', ' ', round(sharpe, 3),
' ', round(maxdrawdown, 3))
plt.legend()
plt.savefig(PATH_prefix + 'backtest.png')
plt.show()
def backtest(agent, env, path, framework):
logger.debug("Backtest")
agents = []
agents.extend(agent)
agents.append(UCRP())
agents.append(Loser())
agents.append(Winner())
labels = [framework, 'UCRP', "Loser", "Winner"]
wealths_result = []
rs_result = []
for i, agent in enumerate(agents):
stocktrader = StockTrader()
info = env.step(None, None, False)
r, done, s, w1, p, risk = parse_info(info)
done = 1
wealth = 10000000
wealths = [wealth]
rs = [1]
while done:
w2 = agent.predict(s, w1)
env_info = env.step(w1, w2, False)
r, done, s_next, w1, p, risk = parse_info(env_info)
wealth = wealth * math.exp(r)
rs.append(math.exp(r) - 1)
wealths.append(wealth)
s = s_next
stocktrader.update_summary(0, r, 0, 0, w2, p)
stocktrader.write(map(lambda x: str(x), env.get_codes()), labels[i])
logger.debug('Finished agents {}'.format(i))
wealths_result.append(wealths)
rs_result.append(rs)
logger.info('资产名称 \t 平均日收益率 \t 夏普率 \t 最大回撤')
plt.figure(figsize=(8, 6), dpi=100)
for i in range(len(agents)):
plt.plot(wealths_result[i], label=labels[i])
mrr = float(np.mean(rs_result[i]) * 100)
sharpe = float(
np.mean(rs_result[i]) / np.std(rs_result[i]) * np.sqrt(252))
maxdrawdown = float(
max(1 - min(wealths_result[i]) /
np.maximum.accumulate(wealths_result[i])))
logger.info("%s \t %s \t %s \t %s", labels[i], round(mrr, 3),
round(sharpe, 3), round(maxdrawdown, 3))
plt.legend()
plt.savefig(path + 'backtest.png')
def backtest(agent,env):
print("starting to backtest......")
from agents.UCRP import UCRP
from agents.Winner import WINNER
from agents.Losser import LOSSER
agents=[]
agents.append(agent)
#agents.append(WINNER())
agents.append(UCRP())
#agents.append(LOSSER())
#labels=['PG','Winner','UCRP','Losser']
labels=['PG','UCRP']
wealths_result=[]
rs_result=[]
for i,agent in enumerate(agents):
info = env.step(None, None)
r, contin, s, w1, p, risk = parse_info(info)
contin = 1
wealth=10000
wealths = [wealth]
rs=[1]
while contin:
w2 = agent.predict(s, w1)
if i==0:
print(w2)
env_info = env.step(w1, w2)
r, contin, s_next, w1, p, risk = parse_info(env_info)
wealth=wealth*math.exp(r)
rs.append(math.exp(r)-1)
wealths.append(wealth)
s=s_next
print('finish one agent')
wealths_result.append(wealths)
rs_result.append(rs)
for i in range(len(agents)):
plt.plot(wealths_result[i],label=labels[i])
print(labels[i],' ',np.mean(rs_result[i]),' ',np.std(rs_result[i]))
plt.legend()
plt.show()
def backtest(pg, env, codes, mode, market, predictor):
global PATH_prefix
agents = []
agents.append(pg)
ucrp = UCRP()
agents.append(ucrp)
mpt = MPT()
agents.append(mpt)
strategy_labels = [predictor, 'UCRP', 'MinVar']
all_growth = pd.DataFrame()
final_info = pd.DataFrame(index=strategy_labels)
for i, strategy in enumerate(agents):
G_t, y_t, price_t, terminal = env.reset()
total_w1 = []
w1_t = torch.zeros(G_t.shape[0])
w1_t[0] = 1
total_w1.append(w1_t)
G_t = torch.Tensor(G_t)
total_reward = torch.tensor(0)
daily_returns = []
cumulative_growth = []
t = 0
while terminal:
if strategy_labels[i] == 'MinVar':
window_returns = price_t[1:, 1:] / price_t[1:, :-1]
w2_t = strategy.forward(window_returns)
w2_t = torch.Tensor(w2_t)
else:
w2_t = strategy.forward(G_t.reshape(-1), w1_t)
mu = cost * (torch.abs(w2_t[1:] - w1_t[1:])).sum()
G_t2, y_t2, price_t2, terminal = env.step()
y_t2 = torch.Tensor(y_t2)
reward = torch.dot(w2_t, y_t2) * (1 - mu)
reward = -torch.log(reward)
w1_t2 = (w2_t * y_t2) / torch.dot(w2_t, y_t2)
total_reward = total_reward + reward
daily_returns.append(torch.exp(-reward) - 1)
cumulative_growth.append(torch.exp(-total_reward))
w1_t = w1_t2
G_t = torch.Tensor(G_t2)
price_t = price_t2
total_w1.append(w1_t)
t = t + 1
SR, exp_return, std_dev = sharpe_ratio(torch.stack(daily_returns))
mdd = max_drawdown(torch.stack(cumulative_growth))
print("Calculating Sharpe Ratio with average daily return of:",
exp_return.item(), ' %', 'and std of:', std_dev.item(), ' %')
print('Sharp ratio for ' + strategy_labels[i] + ':', SR)
print('MDD for ' + strategy_labels[i] + ':', mdd)
print('Total steps per epoch:', t)
print('Final Test Growth for ' + strategy_labels[i] + ':',
cumulative_growth[-1])
print('Final weight allocation for ' + strategy_labels[i] + ':', w1_t)
consolidate_info(final_info, strategy_labels[i], SR.item(),
exp_return.item(), std_dev.item(), mdd.item(),
cumulative_growth[-1].item())
np.savetxt(r'' + PATH_prefix + strategy_labels[i] + '_weights_' + mode,
torch.stack(total_w1).data.numpy(),
delimiter=",")
all_growth[strategy_labels[i]] = torch.stack(
cumulative_growth).data.numpy()
plt.plot(cumulative_growth, label=strategy_labels[i])
all_growth.to_csv(r'' + PATH_prefix + 'all_growth_' + mode + '_rc_' +
str(rc_factor) + '_' + predictor + '_' + market +
'_hs_' + str(pg.n_hidden) + '_back_test.csv')
final_info.to_csv(r'' + PATH_prefix + 'final_info_' + mode + '_rc_' +
str(rc_factor) + '_' + predictor + '_' + market +
'_hs_' + str(pg.n_hidden) + '_back_test.csv')
plt.legend()
plt.savefig(r'' + PATH_prefix + mode + '_rc_' + str(rc_factor) + '_' +
predictor + '_' + market + '_hs_' + str(pg.n_hidden) +
'_back_test.png')
plt.show()
def backtest(agent, env):
global PATH_prefix
print("starting to backtest......")
from agents.UCRP import UCRP
from agents.Winner import WINNER
from agents.Losser import LOSSER
agents = []
agents.extend(agent)
agents.append(WINNER())
agents.append(UCRP())
agents.append(LOSSER())
labels = ['PG', 'Winner', 'UCRP', 'Losser']
wealths_result = []
rs_result = []
for i, agent in enumerate(agents):
stocktrader = StockTrader()
info = env.step(None, None, 'False')
r, contin, s, w1, p, risk = parse_info(info)
contin = 1
wealth = 10000
wealths = [wealth]
rs = [1]
while contin:
w2 = agent.predict(s, w1)
env_info = env.step(w1, w2, 'False')
r, contin, s_next, w1, p, risk = parse_info(env_info)
wealth = wealth * math.exp(r)
rs.append(math.exp(r) - 1)
wealths.append(wealth)
s = s_next
stocktrader.update_summary(0, r, 0, 0, w2, p)
stocktrader.write(map(lambda x: str(x), env.get_codes()), labels[i])
print('finish one agent')
wealths_result.append(wealths)
rs_result.append(rs)
print('资产名称', ' ', '平均日收益率', ' ', '夏普率', ' ', '最大回撤')
plt.figure(figsize=(8, 6), dpi=100)
for i in range(len(agents)):
# if labels[i] == 'UCRP' or labels[i] == 'Losser':
# continue
plt.plot(wealths_result[i], label=labels[i])
fileObject = open(labels[i] + '.txt', 'w')
for ip in wealths_result[i]:
fileObject.write(str(ip))
fileObject.write(', ')
fileObject.close()
mrr = float(np.mean(rs_result[i]) * 100)
sharpe = float(
np.mean(rs_result[i]) / np.std(rs_result[i]) * np.sqrt(252))
maxdrawdown = float(
max(1 - min(wealths_result[i]) /
np.maximum.accumulate(wealths_result[i])))
print(labels[i], ' ', round(mrr, 3), '%', ' ', round(sharpe, 3),
' ', round(maxdrawdown, 3))
# new_data = []
# for i in LSTM:
# for j in range(8):
# new_data.append(i)
# inter_op = interpolate.interp1d(range(len(hand_data)), hand_data, kind='linear')
# hand_data = inter_op(7*range(len(hand_data)))
# def longer(i):
# k = np.zeros(180)
# j = 0
# p = 0
# while j + 1 < len(i):
# print([p, j])
# k[p] = i[j]
# k[p + 1] = i[j]
# k[p + 2] = i[j + 1]
# p += 3
# j += 2
# return i
# LSTM = longer(LSTM)
# Random = longer(Random)
# Uniform = longer(Uniform)
# plt.plot(LSTM, label='LSTM')
# plt.plot(Random, label='Random')
# plt.plot(Uniform, label='Uniform')
plt.legend()
plt.savefig(PATH_prefix + 'backtest_with_hand.png')
plt.show()