blbd4ngge
August 27, 2024
1 How to use the Kalman Filter and Wavelets
1.1 Imports & Settings
[1]: import warnings
[Link]('ignore')
[2]: %matplotlib inline
from datetime import datetime
import itertools
import pandas as pd
import pandas_datareader.data as web
from pykalman import KalmanFilter
import pywt
import [Link] as plt
import seaborn as sns
[3]: [Link]('ignore')
sns.set_style('whitegrid')
idx = [Link]
1.2 Get Data
The assets.h5 store can be generated using the the notebook create_datasets in the data directory
in the root directory of this repo for instruction to download the following dataset.
We load the Quandl stock price datasets covering the US equity markets 2000-18 using
[Link] to perform a slice operation on the [Link], select the adjusted close price
and unpivot the column to convert the DataFrame to wide format with tickers in the columns and
timestamps in the rows:
Set data store location:
[4]: DATA_STORE = '../data/assets.h5'
1
[5]: with [Link](DATA_STORE) as store:
sp500 = store['sp500/stooq'].loc['2009': '2010', 'close']
1.3 Kalman Filter
1.3.1 Configuration
[6]: kf = KalmanFilter(transition_matrices = [1],
observation_matrices = [1],
initial_state_mean = 0,
initial_state_covariance = 1,
observation_covariance=1,
transition_covariance=.01)
1.3.2 Estimate the hidden state
[7]: state_means, _ = [Link](sp500)
1.3.3 Compare with moving average
[8]: sp500_smoothed = sp500.to_frame('close')
sp500_smoothed['Kalman Filter'] = state_means
for months in [1,2,3]:
sp500_smoothed[f'MA ({months}m)'] = [Link](window=months*21).mean()
ax = sp500_smoothed.plot(title='Kalman Filter vs Moving Average',␣
↪figsize=(14,6), lw=1, rot=0)
ax.set_xlabel('')
ax.set_ylabel('S&P 500')
plt.tight_layout()
[Link]();
2
1.4 Wavelets
[9]: wavelet = [Link]('db6')
phi, psi, x = [Link](level=5)
df = [Link]({'$\phi$': phi, '$\psi$': psi}, index=x)
[Link](title='Daubechies', subplots=True, layout=(1, 2), figsize=(14, 4),␣
↪lw=2, rot=0)
plt.tight_layout()
[Link]();
[10]: plot_data = [('db', (4, 3)),
('sym', (4, 3)),
('coif', (3, 2))]
for family, (rows, cols) in plot_data:
fig = [Link](figsize=(24, 12))
fig.subplots_adjust(hspace=0.2, wspace=0.2, bottom=.02, left=.06,
right=.97, top=.94)
colors = [Link]('bgrcmyk')
wnames = [Link](family)
i = iter(wnames)
for col in range(cols):
for row in range(rows):
try:
wavelet = [Link](next(i))
except StopIteration:
break
phi, psi, x = [Link](level=5)
color = next(colors)
3
ax = fig.add_subplot(rows, 2 * cols, 1 + 2 * (col + row * cols))
ax.set_title([Link] + " phi")
[Link](x, phi, color, lw=1)
ax.set_xlim(min(x), max(x))
ax = fig.add_subplot(rows, 2*cols, 1 + 2*(col + row*cols) + 1)
ax.set_title([Link] + " psi")
[Link](x, psi, color, lw=1)
ax.set_xlim(min(x), max(x))
[Link]()
for family, (rows, cols) in [('bior', (4, 3)), ('rbio', (4, 3))]:
fig = [Link](figsize=(24, 12))
fig.subplots_adjust(hspace=0.5, wspace=0.2, bottom=.02, left=.06,
right=.97, top=.94)
colors = [Link]('bgrcmyk')
wnames = [Link](family)
i = iter(wnames)
for col in range(cols):
for row in range(rows):
try:
wavelet = [Link](next(i))
except StopIteration:
break
phi, psi, phi_r, psi_r, x = [Link](level=5)
row *= 2
color = next(colors)
ax = fig.add_subplot(2*rows, 2*cols, 1 + 2*(col + row*cols))
ax.set_title([Link] + " phi")
[Link](x, phi, color, lw=1)
ax.set_xlim(min(x), max(x))
ax = fig.add_subplot(2*rows, 2*cols, 2*(1 + col + row*cols))
ax.set_title([Link] + " psi")
[Link](x, psi, color, lw=1)
ax.set_xlim(min(x), max(x))
row += 1
ax = fig.add_subplot(2*rows, 2*cols, 1 + 2*(col + row*cols))
ax.set_title([Link] + " phi_r")
[Link](x, phi_r, color, lw=1)
ax.set_xlim(min(x), max(x))
ax = fig.add_subplot(2*rows, 2*cols, 1 + 2*(col + row*cols) + 1)
ax.set_title([Link] + " psi_r")
4
[Link](x, psi_r, color, lw=1)
ax.set_xlim(min(x), max(x))
[Link]()
[Link]()
5
6
1.4.1 Available Wavelet Families
[11]: [Link](short=False)
[11]: ['Haar',
'Daubechies',
'Symlets',
'Coiflets',
'Biorthogonal',
'Reverse biorthogonal',
'Discrete Meyer (FIR Approximation)',
'Gaussian',
'Mexican hat wavelet',
'Morlet wavelet',
'Complex Gaussian wavelets',
'Shannon wavelets',
'Frequency B-Spline wavelets',
'Complex Morlet wavelets']
1.4.2 Smoothing the S&P 500
[12]: signal = (pd.read_hdf(DATA_STORE, 'sp500/stooq')
.loc['2008': '2009']
.close.pct_change()
.dropna())
fig, axes = [Link](ncols=2, figsize=(14, 5))
7
wavelet = "db6"
for i, scale in enumerate([.1, .5]):
coefficients = [Link](signal, wavelet, mode='per')
coefficients[1:] = [[Link](i, value=scale*[Link](),␣
↪mode='soft') for i in coefficients[1:]]
reconstructed_signal = [Link](coefficients, wavelet, mode='per')
[Link](color="b", alpha=0.5, label='original signal', lw=2,
title=f'Threshold Scale: {scale:.1f}', ax=axes[i])
[Link](reconstructed_signal, index=[Link]).plot(c='k', label='DWT␣
↪smoothing}', linewidth=1, ax=axes[i])
axes[i].legend()
fig.tight_layout()
[Link]();
