how much are bls estimates out by?

bls.py

import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.patches import Patch
import datetime

# data from https://www.bls.gov/news.release/empsit.nr0.htm
excel_file = '/content/cesvin00.xlsx'
df_data    = pd.read_excel(excel_file, sheet_name='Data', header=2)

# Reference‐months
date_cols  = [c for c in df_data.columns if isinstance(c, str) and '_' in c]

# Grab first & final rows
label_col  = 'Relating to Period'
first_row  = df_data[df_data[label_col]
                     .str.contains("First Preliminary", na=False)].iloc[0]
third_row  = df_data[df_data[label_col]
                     .str.contains("Third Preliminary", na=False)].iloc[0]

# Build the error DataFrame
df = pd.DataFrame({
    'Reference Month': date_cols, # Keep as string initially for manual parsing
    'First'          : first_row [date_cols].astype(float).values,
    'Final'          : third_row [date_cols].astype(float).values
})
df['ErrorPct'] = (df['Final'] - df['First']) / df['First'] * 100

# Manually parse reference-month strings into correct datetimes
def parse_ref(s):
    mon, yy = s.split('_')
    year = int(yy)
    # Years 00–25 → 2000–2025; 26–99 → 1926–1999
    full_year = 2000 + year if year <= 25 else 1900 + year
    month_num = datetime.datetime.strptime(mon, '%b').month
    return datetime.date(full_year, month_num, 1)

df['Reference Month'] = df['Reference Month'].apply(parse_ref)



# — FILTER to June 2003 through last available (Dec 2024) — CHANGE THIS In FUTURE
start, end = datetime.date(2003, 1, 1), datetime.date(2024, 12, 1)
df_plot = df[(df['Reference Month'] >= start) & (df['Reference Month'] <= end)].copy()


import pandas as pd

# First, make sure Reference Month is datetime:
df_plot['Reference Month'] = pd.to_datetime(df_plot['Reference Month'])

# Build a list of the new rows you need:
new_rows = [
    {'Reference Month': '2025-02-01', 'First': 159218.0, 'Final': 159155.0, 'ErrorPct': -0.039568},
    {'Reference Month': '2025-03-01', 'First': 159398.0, 'Final': 159275.0, 'ErrorPct': -0.077165},
    {'Reference Month': '2025-04-01', 'First': 159517.0, 'Final': 159433.0, 'ErrorPct': -0.052659},
    {'Reference Month': '2025-05-01', 'First': 159561.0, 'Final': 159452.0, 'ErrorPct': -0.068312},
    {'Reference Month': '2025-06-01', 'First': 159724.0, 'Final': 159466.0, 'ErrorPct': -0.161529},
]

# Turn them into a DataFrame (ensuring the date column is parsed):
df_new = pd.DataFrame(new_rows)
df_new['Reference Month'] = pd.to_datetime(df_new['Reference Month'])

# Append and then drop any duplicates (keeping the appended rows):
df_plot = pd.concat([df_plot, df_new], ignore_index=True)
df_plot = (
    df_plot
    .drop_duplicates(subset='Reference Month', keep='last')
    .sort_values('Reference Month')
    .reset_index(drop=True)
)

# Verify:
print(df_plot.tail(7))


# Presidential bands (optional)
terms = [
    ('1993-01-20','2001-01-20','Democratic'),
    ('2001-01-20','2009-01-20','Republican'),
    ('2009-01-20','2017-01-20','Democratic'),
    ('2017-01-20','2021-01-20','Republican'),
    ('2021-01-20','2025-01-20','Democratic'),  # Biden’s term extends into 2025
    ('2025-01-20','2026-01-20','Republican'), 
]

# — PLOT —
fig, ax = plt.subplots(figsize=(10,5))

# shade parties
for a,b,party in terms:
    start_dt = pd.to_datetime(a)
    end_dt = pd.to_datetime(b)
    ax.axvspan(start_dt, end_dt, color=('blue' if party=='Democratic' else 'red'),
               alpha=0.1, edgecolor='none')

# plot error
ax.plot(df_plot['Reference Month'], df_plot['ErrorPct'], lw=1)
ax.set_xlim(start, end) # Use datetime.date objects for xlim
ax.set_ylim(-1.2, 1.2)  # Symmetric limits
#ax.set_xlabel('Reference Month')
ax.set_ylabel('Error (% difference)')
ax.set_title('Error Rates in Monthly Payroll Employment Estimates\n'
             '% difference between initial (1st) and final (3rd) estimate')

# Data credit
plt.figtext(
    0.98, 0.01,
    "data:bls.gov by @iamreddave",
    ha="right", va="bottom",
    fontsize=8, alpha=0.7
)


# legend for shading
legend_handles = [
    Patch(facecolor='blue', alpha=0.1, label='Democratic'),
    Patch(facecolor='red', alpha=0.1, label='Republican'),
]
ax.legend(handles=legend_handles, loc='upper left')

fig.autofmt_xdate()
plt.tight_layout()
fig.savefig("payroll_error_rates_2000_2025.png", dpi=300, bbox_inches='tight')

plt.show()