백테스트 - MLP 5분봉

pred.sum()​

# df = pyupbit.get_ohlcv("KRW-BTC", "minute5", count=12*24*30*3)
df = pd.read_csv("20220103_btc_minute5_90days.csv").rename(columns = {"Unnamed: 0":'datetime'}).set_index("datetime")
df = df[['close','volume']]
df

data = df.copy()
data['return'] = np.log(data['close']/data['close'].shift(1))
data.dropna(inplace=True)

data['direction'] = np.where(data['return']>0, 1, 0)

lags = 5
cols = []
for lag in range(1, lags+1):
    col = "lag_{}".format(lag)
    data[col] = data['return'].shift(lag)
    cols.append(col)
data.dropna(inplace=True)
data

 

data['momentum'] = data['return'].rolling(5).mean().shift(1)
data['volatility'] = data['return'].rolling(20).std().shift(1)
data['distance'] = (data['close']-data['close'].rolling(50).mean()).shift(1)
data.dropna(inplace=True)
cols.extend(['momentum','volatility','distance'])

cutoff = '2021-12-01'
training_data = data[data.index<cutoff].copy()
mu, std = training_data.mean(), training_data.std()
training_data_ = (training_data-mu)/std

test_data = data[data.index>=cutoff].copy()
test_data_ = (test_data-mu)/std

import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.optimizers import Adam, RMSprop

optimizer = Adam(learning_rate=0.0001)

def set_seeds(seed=100):
    #random.seed(seed)
    np.random.seed(seed)
    tf.random.set_seed(100)
set_seeds()

model = Sequential()
model.add(Dense(32, activation='relu', 
               input_shape=(len(cols),)))
model.add(Dropout(0.2))
model.add(Dense(32, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(32, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer=optimizer,
             loss='binary_crossentropy',
             metrics=['accuracy'])

%%time 
model.fit(training_data_[cols], training_data['direction'],
         epochs=20, verbose=False,
         validation_split=0.2, shuffle=False)
res = pd.DataFrame(model.history.history)
res[['accuracy','val_accuracy']].plot(figsize=(10,6), style='--')

cutoff_value = 0.5
pred = np.where(model.predict(training_data_[cols])>cutoff_value, 1,0)

training_data['prediction'] = np.where(pred > 0, 1, 0)
training_data['strategy'] = (training_data['prediction']*
                            training_data['return'])
training_data[['return','strategy']].sum().apply(np.exp)

training_data[['return','strategy']].cumsum().apply(np.exp).plot(figsize=(10,6))

 

cutoff_value = 0.5
pred = np.where(model.predict(test_data_[cols])>cutoff_value, 1,0)
pred[:30].flatten()

test_data['prediction'] = np.where(pred > 0, 1, 0)
test_data['strategy'] = (test_data['prediction']*
                            test_data['return'])
test_data[['return','strategy']].sum().apply(np.exp)

test_data[['return','strategy']].cumsum().apply(np.exp).plot(figsize=(10,6))

▶ 테스트 데이터는 하락장을 버티지 못하였다. 수수료가 없음에도 훈련데이터보다 뒤떨어지는 5% 수익률 

 

○ precision으로 평가지표 변경하기

import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.optimizers import Adam, RMSprop
from keras.metrics import Precision

optimizer = Adam(learning_rate=0.0001)

def set_seeds(seed=100):
    #random.seed(seed)
    np.random.seed(seed)
    tf.random.set_seed(100)
set_seeds()

model = Sequential()
model.add(Dense(32, activation='relu', 
               input_shape=(len(cols),)))
model.add(Dropout(0.2))
model.add(Dense(32, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(32, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer=optimizer,
             loss='binary_crossentropy',
             metrics=['Precision'])

%%time 
model.fit(training_data_[cols], training_data['direction'],
         epochs=20, verbose=False,
         validation_split=0.2, shuffle=False)
res = pd.DataFrame(model.history.history)
res[['precision','val_precision']].plot(figsize=(10,6), style='--')

cutoff_value = 0.5
pred = np.where(model.predict(training_data_[cols])>cutoff_value, 1,0)

training_data['prediction'] = np.where(pred > 0, 1, 0)
training_data['strategy'] = (training_data['prediction']*
                            training_data['return'])
training_data[['return','strategy']].sum().apply(np.exp)

training_data[['return','strategy']].cumsum().apply(np.exp).plot(figsize=(10,6))

cutoff_value = 0.5
pred = np.where(model.predict(test_data_[cols])>cutoff_value, 1,0)

test_data['prediction'] = np.where(pred > 0, 1, 0)
test_data['strategy'] = (test_data['prediction']*
                            test_data['return'])
test_data[['return','strategy']].sum().apply(np.exp)

test_data[['return','strategy']].cumsum().apply(np.exp).plot(figsize=(10,6))