forked from AI4Finance-Foundation/FinRL_Crypto
-
Notifications
You must be signed in to change notification settings - Fork 0
/
4_backtest.py
267 lines (215 loc) · 11.1 KB
/
4_backtest.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
"""
This script contains a set of functions for loading and processing data for trading.
It contains the following functions:
load_validated_model: Loads the best trial from the pickle file in the specified directory and returns the best trial's attributes.
download_CVIX: Downloads the CVIX dataframe from Yahoo finance and returns it.
load_and_process_data: loads and process the trade data from the specified data folder and returns the data.
After that, the large loop analyzes every result by creating an instance of an Alpaca environment and checking
what the model would do through the environment using the new trading data
Finally, the resulting backtests are analyzes for performance a performance metric per benchmark (EQW, S&P BCI) plus
all the input DRL agents are analyzed.
"""
import pickle
import matplotlib.dates as mdates
from config_main import *
from function_finance_metrics import *
from processor_Yahoo import Yahoofinance
from environment_Alpaca import CryptoEnvAlpaca
from drl_agents.elegantrl_models import DRLAgent as DRLAgent_erl
def load_validated_model(result):
with open('./train_results/' + result + '/best_trial', 'rb') as handle:
best_trial = pickle.load(handle)
print('BEST TRIAL: ', best_trial.number)
timeframe = best_trial.user_attrs['timeframe']
ticker_list = best_trial.user_attrs['ticker_list']
technical_ind = best_trial.user_attrs['technical_indicator_list']
net_dim = best_trial.params['net_dimension']
model_name = best_trial.user_attrs['model_name']
print('\nMODEL_NAME: ', model_name)
print(best_trial.params)
print(timeframe)
name_test = best_trial.user_attrs['name_test']
env_params = {
"lookback": best_trial.params['lookback'],
"norm_cash": best_trial.params['norm_cash'],
"norm_stocks": best_trial.params['norm_stocks'],
"norm_tech": best_trial.params['norm_tech'],
"norm_reward": best_trial.params['norm_reward'],
"norm_action": best_trial.params['norm_action']
}
return env_params, net_dim, timeframe, ticker_list, technical_ind, name_test, model_name
def download_CVIX(trade_start_date, trade_end_date):
trade_start_date = trade_start_date[:10]
trade_end_date = trade_end_date[:10]
TIME_INTERVAL = '60m'
YahooProcessor = Yahoofinance('yahoofinance', trade_start_date, trade_end_date, TIME_INTERVAL)
CVOL_df = YahooProcessor.download_data(['CVOL-USD'])
CVOL_df.set_index('date', inplace=True)
CVOL_df = CVOL_df.resample('5Min').interpolate(method='linear')
return CVOL_df['close']
def load_and_process_data(TIMEFRAME, trade_start_date, trade_end_date):
data_folder = f'./data/trade_data/{TIMEFRAME}_{str(trade_start_date[2:10])}_{str(trade_end_date[2:10])}'
print(f'\nLOADING DATA FOLDER: {data_folder}\n')
with open(data_folder + '/data_from_processor', 'rb') as handle:
data_from_processor = pickle.load(handle)
with open(data_folder + '/price_array', 'rb') as handle:
price_array = pickle.load(handle)
with open(data_folder + '/tech_array', 'rb') as handle:
tech_array = pickle.load(handle)
with open(data_folder + '/time_array', 'rb') as handle:
time_array = pickle.load(handle)
CVIX_df = download_CVIX(trade_start_date, trade_end_date)
CVIX_df = pd.merge(time_array.to_series(), CVIX_df, left_index=True, right_index=True, how='left')
cvix_array = CVIX_df['close'].values
cvix_array_growth = np.diff(cvix_array)
return data_from_processor, price_array, tech_array, time_array, cvix_array, cvix_array_growth
# Inputs
#######################################################################################################
#######################################################################################################
#######################################################################################################
print('TRADE_START_DATE ', trade_start_date)
print('TRADE_END_DATE ', trade_end_date, '\n')
pickle_results = ["res_2023-01-23__16_32_55_model_WF_ppo_5m_3H_20k",
"res_2023-01-23__17_07_49_model_KCV_ppo_5m_3H_20005k",
"res_2023-01-23__16_44_30_model_CPCV_ppo_5m_3H_20k"
]
# Execution
#######################################################################################################
#######################################################################################################
drl_cumrets_list = []
model_names_list = []
_, _, timeframe, ticker_list, technical_ind, _, _ = load_validated_model(pickle_results[0])
data_from_processor, price_array, tech_array, time_array, cvix_array, cvix_array_growth = load_and_process_data(TIMEFRAME, trade_start_date, trade_end_date)
for count, result in enumerate(pickle_results):
env_params, net_dim, timeframe, ticker_list, technical_ind, name_test, model_name = load_validated_model(result)
model_names_list.append(model_name)
cwd = './train_results/' + result + '/stored_agent/'
data_config = {
"cvix_array": cvix_array,
"cvix_array_growth": cvix_array_growth,
"time_array": time_array,
"price_array": price_array,
"tech_array": tech_array,
"if_train": False,
}
env = CryptoEnvAlpaca
env_instance = env(config=data_config,
env_params=env_params,
if_log=True
)
account_value_erl = DRLAgent_erl.DRL_prediction(
model_name=model_name,
cwd=cwd,
net_dimension=net_dim,
environment=env_instance,
gpu_id=0
)
# Correct slicing (due to DRL start/end)
lookback = env_params['lookback']
indice_start = lookback - 1
indice_end = len(price_array) - lookback
time_array = time_array[indice_start:indice_end]
# Slice cvix array
if count == 0:
cvix_array = cvix_array[indice_start:indice_end]
cvix_array_growth = cvix_array_growth[indice_start:indice_end]
# Compute Sharpe's of each coin
account_value_eqw, ewq_rets, eqw_cumrets = compute_eqw(price_array, indice_start, indice_end)
# Compute annualization factor
data_points_per_year = compute_data_points_per_year(timeframe)
dataset_size = np.shape(ewq_rets)[0]
factor = data_points_per_year / dataset_size
# Compute DRL rets
account_value_erl = np.array(account_value_erl)
drl_rets = account_value_erl[1:] - account_value_erl[:-1]
drl_cumrets = [x / account_value_erl[0] - 1 for x in account_value_erl]
drl_cumrets_list.append(drl_cumrets)
# Compute metrics per pickle result
#######################################################################################################
# Only compute consistent metrics once
if count == 0:
# Load S&P index
spy_index_df = pd.read_csv('data/SPY_Crypto_Broad_Digital_Market_Index - Sheet1.csv')
spy_index_df['Date'] = pd.to_datetime(spy_index_df['Date'])
account_value_spy = np.array(spy_index_df['S&P index'])
spy_rets = account_value_spy[:-1] / account_value_spy[1:] - 1
spy_rets = np.insert(spy_rets, 0, 0)
spy_index_df['cumrets_sp_idx'] = [x / spy_index_df['S&P index'][0] - 1 for x in spy_index_df['S&P index']]
spy_index_df['rets_sp_idx'] = spy_rets
spy_index_df.set_index('Date', inplace=True)
spy_index_df = spy_index_df.resample('5Min').interpolate(method='pchip')
sp_annual_ret, sp_annual_vol, sp_sharpe_rat, sp_vol = aggregate_performance_array(spy_rets, factor)
write_metrics_to_results('S&P Broad Crypto index',
'plots_and_metrics/test_metrics.txt',
spy_index_df['cumrets_sp_idx'],
sp_annual_ret,
sp_annual_vol,
sp_sharpe_rat,
sp_vol,
'w'
)
# Write buy-and-hold strategy
eqw_annual_ret, eqw_annual_vol, eqw_sharpe_rat, eqw_vol = aggregate_performance_array(np.array(ewq_rets),
factor)
write_metrics_to_results('Buy-and-Hold',
'plots_and_metrics/test_metrics.txt',
eqw_cumrets,
eqw_annual_ret,
eqw_annual_vol,
eqw_sharpe_rat,
eqw_vol,
'a'
)
# Then compute the actual metrics from the DRL agents
drl_annual_ret, drl_annual_vol, drl_sharpe_rat, drl_vol = aggregate_performance_array(np.array(drl_rets), factor)
write_metrics_to_results(model_name,
'plots_and_metrics/test_metrics.txt',
drl_cumrets,
drl_annual_ret,
drl_annual_vol,
drl_sharpe_rat,
drl_vol,
'a'
)
# Hold out of loop only add once
#######################################################################################################
# Plot
#######################################################################################################
#######################################################################################################
drl_rets_array = np.transpose(np.vstack(drl_cumrets_list))
# General 1
plt.rcParams.update({'font.size': 22})
plt.figure(dpi=300)
f, ax1 = plt.subplots(figsize=(20, 8))
# Plot returns
line_width = 2
ax1.plot(spy_index_df.index, spy_index_df['cumrets_sp_idx'].values, linewidth=3, label='S&P BDM Index')
ax1.plot(time_array, eqw_cumrets[1:], linewidth=line_width, label='Equal-weight', color='blue')
for i in range(np.shape(drl_rets_array)[1]):
ax1.plot(time_array, drl_rets_array[:, i], label=model_names_list[i], linewidth=line_width)
ax1.legend(frameon=False, ncol=len(model_names_list) + 2, loc='upper left', bbox_to_anchor=(0, 1.11))
ax1.patch.set_edgecolor('black')
ax1.patch.set_linewidth(3)
ax1.grid()
# Plot CVIX
ax2 = ax1.twinx()
ax2.plot(time_array, cvix_array, linewidth=4, label='CVIX', color='black', linestyle='dashed', alpha=0.4)
ax2.legend(frameon=False, loc='upper right', bbox_to_anchor=(0.7, 1.17))
ax2.patch.set_edgecolor('black')
ax2.patch.set_linewidth(3)
ax2.set_ylabel('CVIX')
ax1.xaxis.set_major_formatter(mdates.DateFormatter('%m/%d'))
ax1.xaxis.set_major_locator(mdates.DayLocator(interval=8))
ax1.set_ylabel('Cumulative return')
plt.xlabel('Date')
plt.legend()
plt.savefig('./plots_and_metrics/test_cumulative_return.png', bbox_inches='tight')
ax2.patch.set_edgecolor('black')
ax2.patch.set_linewidth(3)
ax2.set_ylabel('CVIX')
ax1.xaxis.set_major_formatter(mdates.DateFormatter('%m/%d'))
ax1.xaxis.set_major_locator(mdates.DayLocator(interval=8))
ax1.set_ylabel('Cumulative return')
plt.xlabel('Date')
plt.legend()
plt.savefig('./plots_and_metrics/test_derivative_CVIX.png', bbox_inches='tight')