import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
Run to view results
pip install xlrd
Run to view results
pip install openpyxl
Run to view results
import xlrd
import openpyxl
Run to view results
ffr = pd.read_excel('/work/fedfunds.xls')
ffr.head(10)
Run to view results
ffr.dtypes #Check data type
Run to view results
#R star
rff = pd.read_excel('r_star.xlsx')
rff.rename(columns={'rstar': 'RSTAR'}, inplace=True) #Match format and style of column title to other dataframes
rff.head(10)
Run to view results
rff.dtypes #Check data type
Run to view results
#SPF 10 year inflation expectations
spf_10yr_inflation_exp = pd.read_excel('spf_10_yr_inflation_expectation.xlsx')
spf_10yr_inflation_exp.tail(10)
Run to view results
ffr.reset_index(drop=True, inplace=True) #Reset index
ffr['observation_date'] = pd.to_datetime(ffr['observation_date']).dt.to_period("Q") #Convert monthly frequency to quarterly frequency
ffr = pd.DataFrame(ffr.groupby('observation_date')['FEDFUNDS'].mean()) #Take the mean for each quarter similar to the Fed of St. Louis (https://fred.stlouisfed.org/series/FEDFUNDS#0)
ffr.head(10)
Run to view results
rff.reset_index(drop=True, inplace=True) #Reset index
rff['Date'] = pd.to_datetime(rff['Date']).dt.to_period("Q") #Convert monthly frequency to quarterly frequency
rff.head(10)
Run to view results
spf_10yr_inflation_exp['Date'] = spf_10yr_inflation_exp['YEAR'].astype(str) + '-Q' + spf_10yr_inflation_exp['QUARTER'].astype(str) #set one column for the date
spf_10yr_inflation_exp['Date'] = pd.PeriodIndex(spf_10yr_inflation_exp['Date'], freq='Q') #Set date in quarter frequency
spf_10yr_inflation_exp.drop(['YEAR', 'QUARTER'], axis=1, inplace=True) #Drop original columns for cleanliness
# Filter the DataFrame based on the condition
spf_10yr_inflation_exp = spf_10yr_inflation_exp[spf_10yr_inflation_exp["Date"] >= "1991Q4"]
spf_10yr_inflation_exp.head(10)
Run to view results
first_non_empty_row_index = spf_10yr_inflation_exp["INFCPI10YR"].first_valid_index()
first_non_empty_row = spf_10yr_inflation_exp.iloc[first_non_empty_row_index,:]
print(f"First non-empty row index: \n{first_non_empty_row}")
Run to view results
ff_star = pd.merge(spf_10yr_inflation_exp, rff, left_on='Date', right_on='Date', how='inner')
ff_star["FFSTAR"] = ff_star["RSTAR"] + ff_star["INFCPI10YR"] #Compute ff*
ff_star.head(10)
Run to view results
mp_stance_df = pd.merge(ff_star, ffr, left_on="Date", right_on="observation_date", how='left')
mp_stance_df.head(10)
Run to view results
mp_stance_df["MP"] = mp_stance_df["FEDFUNDS"] - mp_stance_df["FFSTAR"] #Use formula
mp_stance_df.head(10)
Run to view results
#Line chart drops the quarterly format. So, we copy the current dataset and adapt it for the visual representation
mp_stance_df_graph = mp_stance_df.copy()
mp_stance_df_graph['Date'] = mp_stance_df_graph['Date'].dt.to_timestamp()
mp_stance_df_graph.head(10)
Run to view results
sns.set(style='ticks')
fig, ax = plt.subplots(figsize=(10, 6))
sns.lineplot(data = mp_stance_df_graph, x="Date", y='MP', ax = ax, linewidth=2)
ax.axhline(0, color='black', linewidth=1.5, linestyle='--')
ax.yaxis.set_tick_params(width=1.5, color='black')
ax.set_title("A Measure of Stance of Monetary Policy")
ax.set_xlabel('Date')
ax.set_ylabel('Percentage (%)')
fig.tight_layout()
plt.show()
Run to view results
factor_premia = pd.read_excel('Century of Factor Premia Monthly.xlsx')
factor_premia.head(30)
Run to view results
#Fix the shape of the dataframe
#Set columns
factor_premia.columns = factor_premia.iloc[17] #Column headers
clean_factor_premia = factor_premia[18:]
clean_factor_premia.tail(10)
Run to view results
clean_factor_premia.dtypes
Run to view results
for column in clean_factor_premia.columns[1:]:
clean_factor_premia[column] = clean_factor_premia[column].astype(float)
clean_factor_premia.dtypes
Run to view results
Now, X-variables have float type.
Since X variables have a monthly frequency and y variable has a quarterly frequency, X variables will have a repeated value for each quarter. To deal with this issue, customizable resampling techniques are applied based on the skewness of the distribution of each factor.
clean_factor_premia = clean_factor_premia.rename(columns={clean_factor_premia.columns[0]: "Date"})
clean_factor_premia['Date'] = pd.to_datetime(clean_factor_premia['Date'])
# Set 'Date' column as the index
clean_factor_premia.set_index('Date', inplace=True)
# Define the resampling methods based on distribution
resampling_methods = {}
# Iterate over the columns
for column in clean_factor_premia.iloc[:, 1:]:
data = clean_factor_premia[column]
# Check the skewness of the data
skewness = data.skew()
# Determine the appropriate resampling method based on skewness
if np.abs(skewness) < 1:
resampling_methods[column] = 'mean'
elif skewness >= 1:
resampling_methods[column] = 'first'
else:
resampling_methods[column] = 'last'
# Create an empty DataFrame to store the resampled data
resampled_data = pd.DataFrame()
# Iterate over the columns and apply the corresponding resampling method
for column, method in resampling_methods.items():
resampled_data[column] = clean_factor_premia[column].resample('Q').apply(method)
resampled_data.head(10)
Run to view results
# Assuming the 'Date' column is already set as the index in resampled_data DataFrame
# Resample the index from monthly to quarterly
resampled_data.index = pd.PeriodIndex(resampled_data.index, freq='Q')
resampled_data.head()
Run to view results
full_df = pd.merge(mp_stance_df, resampled_data, left_on="Date", right_on=resampled_data.index, how='inner')
full_df.set_index(full_df["Date"], inplace=True)
full_df = full_df.drop(['INFCPI10YR', 'RSTAR', 'FFSTAR', 'FEDFUNDS', 'Date_x', 'Date_y', 'Date'], axis=1) #Drop irrelevant columns
full_df.head(10)
Run to view results
# Filter and retrieve rows where "Date" is NaN
nan_date_rows = full_df[full_df.index.isna()]
# Print the rows where "Date" is NaN
print(nan_date_rows)
Run to view results
It does not. SO, we can check that variables have appropriate data tyep.
full_df.dtypes
Run to view results
y = full_df["MP"]
X = full_df.iloc[:, 1:]
Run to view results
# Add a constant column to X
X = sm.add_constant(X)
# Fit the OLS model
model = sm.OLS(y, X)
results = model.fit()
# Print the summary statistics
print(results.summary())
Run to view results
import pandas as pd
import numpy as np
from sklearn.linear_model import LinearRegression
from mlxtend.feature_selection import SequentialFeatureSelector
# Assuming X and y are your predictor and target variables, respectively
X = sm.add_constant(X)
# Create an instance of the scikit-learn LinearRegression estimator
estimator = LinearRegression()
# Perform forward stepwise regression
stepwise_selector = SequentialFeatureSelector(estimator=estimator,
k_features='best',
forward=True,
scoring='r2',
cv=0)
stepwise_selector.fit(X, y)
selected_features = list(stepwise_selector.k_feature_names_)
dropped_features = list(set(X.columns) - set(selected_features))
# Fit the final model with selected features
final_model = sm.OLS(y, X[selected_features]).fit()
# Get the summary of the final model
model_summary = final_model.summary()
print(model_summary)
Run to view results
# Print the dropped features, if any
if dropped_features:
print(f"Dropped features: {dropped_features}.")
else:
print(f"No features were dropped during stepwise regression.")
Run to view results
import pandas as pd
from statsmodels.tsa.stattools import grangercausalitytests
def find_best_lags(x_columns, y_variable, data_frame):
results = []
for column in x_columns:
best_lag = None
best_p_value = float('inf')
x_values = data_frame[column].values
y_values = data_frame[y_variable].values
for lag in range(1, 13):
test_data = pd.DataFrame({'x': x_values, 'y': y_values})
test_data = test_data.dropna() # Remove rows with missing values
# Perform Granger causality test for the given lag
results_dict = grangercausalitytests(test_data, maxlag=lag, verbose=False)
p_value = results_dict[lag][0]['params_ftest'][1]
if p_value < best_p_value:
best_lag = lag
best_p_value = p_value
results.append({'Column': column, 'Best Lag': best_lag})
results_df = pd.DataFrame(results)
return results_df
Run to view results
# Example usage
x_columns = [col for col in full_df.columns if col != "MP"] # Replace with the names of the columns you want to test
y_variable = 'MP' # Replace with the name of the dependent variable
lag_results = find_best_lags(x_columns, y_variable, full_df)
print(lag_results)
Run to view results
import pandas as pd
from statsmodels.tsa.stattools import grangercausalitytests
def create_lagged_dataframe(x_columns, y_variable, data_frame):
results = []
for column in x_columns:
best_lag = None
best_p_value = float('inf')
x_values = data_frame[column].values
y_values = data_frame[y_variable].values
test_data = pd.DataFrame({'x': x_values, 'y': y_values})
test_data = test_data.dropna() # Remove rows with missing values
# Perform Granger causality test with the maximum lag value as 12
results_dict = grangercausalitytests(test_data, maxlag=12, verbose=False)
for lag, result in results_dict.items():
p_value = result[0]['params_ftest'][1]
if p_value < best_p_value:
best_lag = lag
best_p_value = p_value
results.append({'Column': column, 'Best_Lag': best_lag})
results_df = pd.DataFrame(results)
# Create lagged DataFrame
lagged_df = pd.DataFrame()
for row in results_df.itertuples():
column_name = row.Column
lag = row.Best_Lag
lagged_df[f'{column_name}_lag{lag}'] = data_frame[column_name].shift(lag)
return lagged_df
Run to view results
# Example usage
x_columns = [col for col in full_df.columns if col != "MP"] # Replace with the names of the columns you want to test
y_variable = 'MP' # Replace with the name of the dependent variable
# Example usage
x_columns = [col for col in full_df.columns if col != "MP"] # Replace with the names of the columns you want to test
y_variable = 'MP' # Replace with the name of the dependent variable
lagged_dataframe = create_lagged_dataframe(x_columns, y_variable, full_df)
Run to view results
lagged_dataframe
Run to view results
import pandas as pd
# Assuming lag_results is a DataFrame containing the lag results
# Assuming lagged_dataframe is your DataFrame
# Find the largest lag value in the lag_results table
largest_lag = lag_results['Best Lag'].max()
# Drop rows from lagged_dataframe based on the largest lag value
lagged_dataframe = lagged_dataframe.iloc[largest_lag:]
# The lagged_dataframe now contains the rows after dropping the largest lag value
Run to view results
no_na_in_first_row = lagged_dataframe.iloc[0, :].isnull().sum() == 0
# Print the result
if no_na_in_first_row:
print("No NaN values in the first row after dropping the largest lag value.")
else:
print("NaN values present in the first row after dropping the largest lag value.")
Run to view results
lag_df = pd.merge(full_df["MP"], lagged_dataframe, left_on=full_df.index, right_on = lagged_dataframe.index, how="inner")
lag_df = lag_df.drop(['key_0'], axis=1)
lag_df
Run to view results
y = lag_df["MP"]
X = lag_df.iloc[:, 1:]
Run to view results
# Add a constant column to X
X = sm.add_constant(X)
# Fit the OLS model
model = sm.OLS(y, X)
results = model.fit()
# Print the summary statistics
print(results.summary())
Run to view results
import pandas as pd
import numpy as np
from sklearn.linear_model import LinearRegression
from mlxtend.feature_selection import SequentialFeatureSelector
# Assuming X and y are your predictor and target variables, respectively
X = sm.add_constant(X)
# Create an instance of the scikit-learn LinearRegression estimator
estimator = LinearRegression()
# Perform forward stepwise regression
stepwise_selector = SequentialFeatureSelector(estimator=estimator,
k_features='best',
forward=True,
scoring='r2',
cv=0)
stepwise_selector.fit(X, y)
selected_features = list(stepwise_selector.k_feature_names_)
dropped_features = list(set(X.columns) - set(selected_features))
# Fit the final model with selected features
final_model = sm.OLS(y, X[selected_features]).fit()
# Get the summary of the final model
model_summary = final_model.summary()
print(model_summary)
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results
Run to view results