def assets(pairs):
fig, ax = plt.subplots()
fig.set_tight_layout(True)
for pair in pairs:
df = priceof(pair).iloc[::60].iloc[-10000:]
df.open = df.open / df.open[0] # normalize to be comparable
dates = df.index.strftime('%Y-%m-%d')
ax.plot(
dates,
df.open,
# marker='.', markersize=1,
ls='-',
lw=1,
label=pair,
)
ax.set_title('Normalized evolution of assets compared to bitcoin')
ax.tick_params(axis='x', labelrotation=40)
ax.xaxis.set_major_locator(ticker.AutoLocator())
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.legend()
return fig
def corrmap():
curves = [key for key in REGISTRY if key.endswith('usdt')]
dfs = []
for curve in curves:
df = priceof(curve)[['open']].rename(columns={'open': curve})
print(curve, df.index[0])
if df.index[0] < datetime(2019, 6, 1):
dfs.append(df)
df = dfs.pop().join(dfs)
corr = df.corr()
sns.heatmap(
corr,
cmap=sns.diverging_palette(20, 220, n=200),
square=True,
)
plt.show()
def _():
from read import priceof
from arsim import AR
past = 200
pos = np.random.randint(400)
btcc = priceof('btcusdt').open.iloc[::60 * 24]
btc = btcc.pct_change().to_numpy()[-past - pos:-pos]
btc_p = btc #btcc.pct_change().to_numpy()[-pos:-pos+past]
p, q = 10, 10
pp, qq = 10, 10
par, A, y = fitarma(btc, p=p, q=q)
# _, Ap, yp = fitarma(btc_p, p=p, q=q)
Ap, yp = A, y
pred = A @ par
predp = Ap @ par
res = y - pred
resp = yp - predp
pgarch, B, u = mom2_garch(res, p=pp, q=qq)
# _, Bp, up = mom2_garch(resp, p=pp, q=qq)
Bp, up = B, u
vol = np.sqrt(B @ pgarch)
volp = np.sqrt(Bp @ pgarch)
plt.plot(np.arange(len(btc_p)), btc_p, label='btc')
plt.plot(np.arange(p + max(p, q), len(btc_p)), predp, label='predictions')
plt.plot(np.arange(p + max(p, q) + pp + qq, len(btc_p)),
volp,
label='volatility')
rvol = pd.Series(resp).rolling(20).std()
plt.plot(np.arange(len(btc_p) - len(rvol), len(btc_p)),
rvol,
label='rolling standard deviation')
plt.legend()
plt.show()
import numpy as np
from read import priceof
def mat(seq, p):
shifts = [seq[p-i-1:-i-1] for i in range(p)]
return np.vstack((*shifts, np.ones(len(seq) - p))).T
def ar(seq, p=1):
return np.linalg.lstsq(mat(seq, p), seq[p:], rcond=None)[0]
returns = priceof('btcusdt').open.iloc[::60*24].pct_change()
params = ar(returns.to_numpy()[-300:], p=2)
residuals = returns[2:] - mat(returns, 2) @ params[:, np.newaxis]
plt.plot(residuals)
plt.show()
plots = [
returns, autocorrelation, autocorrelation_squared, leverage, fat_tailed
]
for pair in [
'btcusdt',
'xrpusdt',
'^TCFA',
'^TCDPIA',
'^DJI',
'^FCHI',
'bnbusdt',
'zrxusdt',
]:
df = priceof(pair)
print(df.columns)
for plot in plots:
for interval in ('minute', 'hourly', 'daily'):
try:
fig = plot(df, interval=interval)
except ValueError:
continue
fig.savefig(
os.path.join(PLOTDIR,
f'{pair}-{plot.__name__}-{interval}.png'))
plt.close(fig)
from read import priceof
import numpy as np
ccs = (
'^TCFA', '^FCHI', '^DJI',
'btcusdt', 'xrpusdt', 'ethusdt',
)
toks = (
'^TCDPIA', 'repbtc', 'batbtc', 'enjbtc', 'omgbtc',
)
for pair in toks:
asset = priceof(pair)
scales = (1,)
if pair not in ('^FCHI', '^DJI'):
scales = np.cumprod((1, 60, 24))
for scale in scales:
returns = asset.open.iloc[::scale].pct_change()
kurt = returns.kurtosis()
print(f'{pair} fisher kurtosis, returns interval {scale}: {kurt}')