def tba(target_quote, alpha=0.05, bounds=(-5, 5)):
ba_ratio = np.log(target_quote['bv1'] / target_quote['av1'])
clipped_ba_ratio = np.clip(ba_ratio, bounds[0], bounds[1]).values
ba_ratio_ema = nbr.numba_ewma(clipped_ba_ratio,
alpha=alpha,
state=None,
adjust=True,
ignore_na=True,
minp=1)[0]
return ba_ratio_ema
def dbook(target_quote, n=1, alpha=0.05):
bv1, av1 = target_quote[['bv1', 'av1']].values.astype(np.float).T
diff_bv1 = np.insert(np.sign(np.diff(bv1, n=n)), [0] * n, 0)
diff_av1 = np.insert(np.sign(np.diff(av1, n=n)), [0] * n, 0)
rs = nbr.numba_ewma((diff_bv1 - diff_av1),
alpha=alpha,
state=None,
adjust=True,
ignore_na=True,
minp=1)[0]
return rs
def tba(target_quote, alpha=0.05):
# ba_ratio = np.log(target_quote['bv1'] / target_quote['av1'])
bv1, av1 = target_quote[['bv1', 'av1']].values.astype(np.float).T
ba_ratio = np.sqrt(
np.divide(bv1, av1, out=np.full_like(bv1, np.nan), where=av1 != 0))
ba_ratio_ema = nbr.numba_ewma(ba_ratio,
alpha=alpha,
state=None,
adjust=True,
ignore_na=True,
minp=1)[0]
return ba_ratio_ema
def calc_absr(df, window):
"""
Estimate the active trading volume (absolute buy sell ratio).
ask, bid volume and amount must be float.
"""
ask, bid, volume, amount = df[['ap1', 'bp1', 'volume',
'turnover']].values.T
vol = volume.copy().astype(float)
np.place(vol, vol == 0, np.nan)
avgcost = np.divide(amount, vol)
spread = ask - bid
# Seperate
avgc_over_ask = avgcost[1:] > ask[:-1]
avgc_below_bid = avgcost[1:] < bid[:-1]
avgc_between = (~avgc_over_ask) & (~avgc_below_bid)
vol_1 = vol[1:]
# calculate active buyer
actibuy = vol_1 * avgc_over_ask
ab_ratios = np.divide((avgcost[1:] - bid[:-1]), spread[:-1])
actibuy[avgc_between] = vol_1[avgc_between] * ab_ratios[avgc_between]
# calculate active seller
actisell = vol_1 * avgc_below_bid
as_ratios = 1 - ab_ratios
actisell[avgc_between] = vol_1[avgc_between] * as_ratios[avgc_between]
actibuy_end = nbr.numba_ewma(actibuy, alpha=0.05)[0][-1]
actisell_end = np.max([nbr.numba_ewma(actisell, alpha=0.05)[0][-1], 1])
out = actibuy_end / actisell_end
# active_volume = np.vstack([actibuy, actisell]).T.astype('float64')
# active_volume = np.nan_to_num(active_volume)
# active_volume_arr = np.full((len(vol), 2), np.nan)
# active_volume_rolsum = np.sum(lut.sliding_window_view(active_volume.T, window), axis=2).T
# active_volume_arr[window:] = active_volume_rolsum
# out = np.log(np.divide(active_volume_arr[:, 0], active_volume_arr[:, 1], out=np.full(len(active_volume_arr), np.nan), where=active_volume_arr[:, 1]!=0))
return out
def calc_factor1(df):
# 基于订单簿的筹码倾向：以价格为标尺
_obp = np.ravel(df.filter(regex=r'[ab]p\d+'))
_obv = np.ravel(df.filter(regex=r'[ab]v\d+'))
# ob_agg = pd.Series({p: np.mean(_obv[_obp == p]) for p in np.unique(_obp) if p != 0})
ob_agg = pd.Series({
p: nbr.numba_ewma(_obv[_obp == p], alpha=0.05)[-1][0]
for p in np.unique(_obp) if p != 0
})
_mid = df['new_price'].iloc[-1]
over = ob_agg[ob_agg.index > _mid].copy()
wgt_over_sum = np.sum(over * over.index / _mid)
under = ob_agg[ob_agg.index < _mid].copy()
# wgt_under_sum = np.sum(under * _mid / under.index)
wgt_under_sum = np.max([np.sum(under * _mid / under.index), 1])
order_ratio = wgt_over_sum / wgt_under_sum
return order_ratio