def strategy(price_data,
signal_a,
signal_b,
window_1=24 * 7,
window_2=24 * 7,
buy_sell_fee=0.0075,
pnl_0=1.0):
sent_score = ah.nb_calc_sentiment_score_b(signal_a, signal_b, window_1,
window_2)
pnl = ah.nb_backtest_a(price_data, sent_score, 1.0, buy_sell_fee)
return pnl
#%%
# define the window size for the sentiment score calculation
n_days = 7
window_size = 24 * n_days # in hours
# generate the sentiment score
sent_score = ah.nb_calc_sentiment_score_a(aug_signal_bullish,
aug_signal_bearish, window_size,
window_size)
# define some parameters for the backtest
start_pnl = 1.0
buy_sell_fee = 0.0075
# run the backtest
pnl = ah.nb_backtest_a(price_data, sent_score, start_pnl, buy_sell_fee)
#%% [markdown]
# # Compare various windows sizes
#%%
sent_score = ah.nb_calc_sentiment_score_a(aug_signal_bullish,
aug_signal_bearish, 1, 2)
pnl = ah.nb_backtest_a(price_data, sent_score, 1.0, 0.0075)
#%%
sent_score = ah.nb_calc_sentiment_score_a(aug_signal_bullish,
aug_signal_bearish, 7 * 24, 7 * 24)
pnl = ah.nb_backtest_a(price_data, sent_score, 1.0, 0.0075)
#%%
# get the signals we're interested in
aug_signal_a = aug_data[:, aug_topics_inv["Positive"]].astype(np.float64)
aug_signal_b = aug_data[:, aug_topics_inv["Bearish"]].astype(np.float64)
sent_score = ah.nb_calc_sentiment_score_c(aug_signal_a, aug_signal_b, 28 * 24,
14 * 24)
date_time = np.array(
[datetime.datetime.utcfromtimestamp(t).isoformat() for t in t_price_data])
# define some parameters for the backtest
start_pnl = 1.0
buy_sell_fee = 0.0075
# run the backtest
pnl = ah.nb_backtest_a(price_data, sent_score, start_pnl, buy_sell_fee)
# set up the figure
fig, ax = plt.subplots(3, 1, sharex=True, sharey=False)
# initialise some labels for the plot
datenum_aug_data = [
md.date2num(datetime.datetime.fromtimestamp(el)) for el in t_aug_data
]
datenum_price_data = [
md.date2num(datetime.datetime.fromtimestamp(el)) for el in t_price_data
]
# plot stuff
ax[0].grid(linewidth=0.4)
ax[1].grid(linewidth=0.4)
ax[2].grid(linewidth=0.4)
ax[0].plot(datenum_price_data, price_data, linewidth=0.5)
ax[1].plot(datenum_aug_data, sent_score, linewidth=0.5)
# It will give 8649 NLP values calculated from 2017 until the beginning of 2019
#%%
#%%
# for each combination of signals, generate PNL for the last period in data
total = np.zeros(shape=(93, 93))
print("calculating... might take a minute or two...")
for i in range(0, len(all_topics)):
for j in range(0, len(all_topics)):
sent_score = ah.nb_calc_sentiment_score_b(all_topics[i],
all_topics[j],
ra_win_size_short=24 * 7,
ra_win_size_long=24 * 7)
pnl = ah.nb_backtest_a(price_data,
sent_score,
1.0,
buy_sell_fee=0.0075)
total[i][j] = pnl[-1]
#print("Row " + str(i+1) + " out of 93...")
print("done")
#%%
#%% [markdown]
# ### Impossible to see all 8649 values
# Chose top 30
#%%
# get all PNL in a dataframe
data = pd.DataFrame(total).rename(columns=aug_topics, index=aug_topics)
# given all combinations of signals, show the combinations that yield the highest PNL
# PNL of various moving window size for a given combination of topics
#%% [markdown]
# # Test strategies with adding noise to moving windows
#%%
sensit = 0.01
aug_signal_a = aug_data[:, aug_topics_inv["Positive"]].astype(np.float64)
aug_signal_b = aug_data[:, aug_topics_inv["Bearish"]].astype(np.float64)
short_win = 26 * 24
long_win = 7 * 24
sent_score = ah.nb_calc_sentiment_score_rand_a(aug_signal_a, aug_signal_b,
short_win, long_win, sensit)
a = ah.nb_backtest_a(price_data, sent_score, start_pnl=1, buy_sell_fee=0.0075)
for i in range(1, 10):
short_win_ran = np.int64(
np.round(np.random.normal(short_win,
np.float64(short_win) * 0.01)))
long_win_ran = np.int64(
np.round(np.random.normal(long_win,
np.float64(long_win) * 0.01)))
#short_win_ran, long_win_ran = short_win, long_win
sent_score = ah.nb_calc_sentiment_score_rand_a(aug_signal_a, aug_signal_b,
short_win_ran, long_win_ran,
sensit)
cum_ret = ah.nb_backtest_a(price_data,
sent_score,
start_pnl=1,