英文:
How to use statsmodels' DynamicFactor method with exogenous variables?
问题
I have a multivariate dynamic factor model with one common factor that I want to estimate with statsmodels.tsa.statespace.dynamic_factor.DynamicFactor
.
The model looks as follows:
As you can see, I am dealing with a t x 4 matrix of endogenous variables. Each of them has 6 own specific exogenous variables, which they don't share. So the only thing the 4 time series have in common is the common factor.
My question is how to put this in code.
I have attempted the following:
model = DynamicFactor(
endog=y, # nobs x 4
exog=X, # nobs x k_exog
k_factors=1,
factor_order=1,
error_order=0,
error_cov_type='diagonal'
)
But the results seem off, and I know from the documentation that X should have the shape of t x k_exog. I am wondering what k_exog should be in my case, and if I can arrange my matrix so that y_1 only uses W_1, etc.
*EDIT: In the model formulation, at one point the dependent variable is called 'NG,' but it should be 'y.' Apologies.
英文:
I have a multivariate dynamic factor model with one common factor that I want to estimate with statsmodels.tsa.statespace.dynamic_factor.DynamicFactor.
The model looks as follows:
Model formulation in LaTeX.*
As you can see, I am dealing with a t x 4 matrix of endogenous variables. Each of them has 6 own specific exogenous variables, which they don't share. So the only thing the 4 time series have in common, is the common factor.
My question is how to put this in code.
I have attempted the following:
model = DynamicFactor(
endog=y, # nobs x 4
exog=X, # nobs x k_exog
k_factors=1,
factor_order=1,
error_order=0,
error_cov_type='diagonal'
)
But the results seem off, and I know from the documentation that X should have the shape of t x k_exog. I am wondering what k_exog should be in my case, and if I can arange my matrix so that y_1 only uses W_1 etc.
*EDIT: in the model formulation, at one point the dependent variable is called 'NG' but it should be y. Apologies.
答案1
得分: 0
DynamicFactor
模型假设每个exog
变量影响每个endog
变量。但是,您可以告诉模型将某些参数的值设置为固定值(而不是估计它们)。您可以使用这个来实现您想要的目标。
一个简单的示例如下:
import numpy as np
import pandas as pd
import statsmodels.api as sm
# 模拟一些数据
nobs = 100
np.random.seed(1234)
y = pd.DataFrame(np.random.normal(size=(nobs, 2)), columns=['y1', 'y2'])
X_1 = pd.Series(np.random.normal(size=nobs), name='x1')
X_2 = pd.Series(np.random.normal(size=nobs), name='x2')
X = pd.concat([X_1, X_2], axis=1)
# 构建模型
mod = sm.tsa.DynamicFactor(y, exog=X, k_factors=1, factor_order=1)
# 如果需要确定需要设置为0的参数的名称,可以打印参数名称
# print(mod.param_names)
# 使用`fix_params`固定适用的参数...
with mod.fix_params({'beta.x2.y1': 0, 'beta.x1.y2': 0}):
# 然后使用`fit`估计其他参数
res = mod.fit(disp=False)
# 打印结果
print(res.summary())
这将产生上述输出结果。
英文:
The DynamicFactor
model assumes that every exog
variable affects every endog
variable. However, you can tell the model to set the values of certain parameters to fixed values (rather than estimate them). You can use this to do what you want.
A simple example follows:
import numpy as np
import pandas as pd
import statsmodels.api as sm
# Simulate some data
nobs = 100
np.random.seed(1234)
y = pd.DataFrame(np.random.normal(size=(nobs, 2)), columns=['y1', 'y2'])
X_1 = pd.Series(np.random.normal(size=nobs), name='x1')
X_2 = pd.Series(np.random.normal(size=nobs), name='x2')
X = pd.concat([X_1, X_2], axis=1)
# Construct the model
mod = sm.tsa.DynamicFactor(y, exog=X, k_factors=1, factor_order=1)
# You can print the parameter names if you need to determine the
# names of the parameters that you need to set fixed to 0
# print(mod.param_names)
# Fix the applicable parameters with `fix_params`...
with mod.fix_params({'beta.x2.y1': 0, 'beta.x1.y2': 0}):
# And estimate the other parameters with `fit`
res = mod.fit(disp=False)
# Print the results
print(res.summary())
Which gives:
Statespace Model Results
=============================================================================================
Dep. Variable: ['y1', 'y2'] No. Observations: 100
Model: DynamicFactor(factors=1, order=1) Log Likelihood -276.575
+ 2 regressors AIC 567.150
Date: Fri, 12 May 2023 BIC 585.386
Time: 22:45:16 HQIC 574.530
Sample: 0
- 100
Covariance Type: opg
===================================================================================
Ljung-Box (L1) (Q): 0.01, 0.10 Jarque-Bera (JB): 5.22, 0.78
Prob(Q): 0.93, 0.76 Prob(JB): 0.07, 0.68
Heteroskedasticity (H): 2.11, 0.75 Skew: -0.56, -0.17
Prob(H) (two-sided): 0.04, 0.41 Kurtosis: 3.02, 3.26
Results for equation y1
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
loading.f1 -0.4278 0.891 -0.480 0.631 -2.174 1.318
beta.x1 -0.0614 0.129 -0.478 0.633 -0.313 0.191
beta.x2 0 nan nan nan nan nan
Results for equation y2
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
loading.f1 0.4487 0.888 0.505 0.613 -1.292 2.189
beta.x1 0 nan nan nan nan nan
beta.x2 -0.1626 0.113 -1.442 0.149 -0.384 0.058
Results for factor equation f1
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
L1.f1 0.1060 0.323 0.328 0.743 -0.527 0.739
Error covariance matrix
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
sigma2.y1 0.6124 0.766 0.800 0.424 -0.889 2.113
sigma2.y2 0.9306 0.818 1.138 0.255 -0.672 2.533
==============================================================================
Warnings:
[1] Covariance matrix calculated using the outer product of gradients (complex-step).
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库,让每个人都能够通过互相帮助和分享经验来进步。
评论