The March employment situation report drops on April 3, 2026, and the question on everyone’s mind is whether February’s payroll miss was a blip or the start of something worse. Weekly unemployment claims, the closest thing to a real-time pulse on the labor market, offer the clearest preview we have. Initial claims for the week ending March 21 came in at 210,000, with continuing claims falling to 1,819,000, the lowest level since May 25, 2024. The 4-week moving average for initial claims sits at 210,500, comfortably below the 223,149 average since 2023. The signal heading into Friday is unambiguous: whatever is happening in the labor market, it is not layoffs.
For context on the employer sentiment side of this story, the March 22 BTOS analysis showed businesses pulling back on hiring while holding steady on overall performance. That framing still holds. This post focuses on what the claims data adds to that picture.
The Python code used to generate each chart is included in this post. Click on any code block to see the full implementation. The complete data pipeline includes:
01_clean_data.py - Processes FRED initial and continuing claims CSVs02_compute_stats.py - Computes summary statistics for inline valuesInitial claims (ICSA) and continuing claims (CCSA) sourced from FRED. DOL weekly release dated March 26, 2026.
DOL Weekly Claims Report | Week ending March 21 | Source: U.S. Department of Labor via FRED
A single week’s claims number is noisy. The 4-week moving average strips out that volatility and reveals the underlying trend. After a brief spike to 230,000 in late January and early February, the moving average has declined steadily and now sits at 210,500. The two panels below tell the story at different scales: the full post-pandemic normalization on the left, and a six-month close-up on the right that makes the recent movements readable.
# =============================================================================
# Initial Claims: Side-by-Side View (Full History + Recent Zoom)
# Source: data/clean/initial_claims.csv (FRED series ICSA)
# =============================================================================
fig, (ax_left, ax_right) = plt.subplots(1, 2, figsize=(10, 4.5),
gridspec_kw={'width_ratios': [1, 1]})
# ── Left panel: Full history ─────────────────────────────────────────────────
ax_left.plot(df_icsa["date"], df_icsa["initial_claims"] / 1000,
color=COLORS['light'], linewidth=0.8, alpha=0.6, label='Weekly')
ax_left.plot(df_icsa["date"], df_icsa["initial_claims_4wma"] / 1000,
color=COLORS['primary'], linewidth=2.5, label='4-Week MA')
# Shade the since-2023 range
ax_left.axhspan(stats['icsa_min_since_2023'] / 1000,
stats['icsa_max_since_2023'] / 1000,
alpha=0.06, color=COLORS['primary'],
label=f"2023-26 range ({stats['icsa_min_since_2023']//1000}K-{stats['icsa_max_since_2023']//1000}K)")
ax_left.set_ylabel("Initial Claims (thousands)")
ax_left.xaxis.set_major_formatter(mdates.DateFormatter('%b\n%Y'))
ax_left.xaxis.set_major_locator(mdates.MonthLocator(interval=9))
ax_left.yaxis.set_major_formatter(ticker.FuncFormatter(lambda x, _: f"{x:.0f}K"))
ax_left.legend(loc='upper right', fontsize=8, framealpha=0.9)
ax_left.set_title("Post-Pandemic Trend",
fontsize=12, fontweight='bold', pad=10)
# ── Right panel: Recent 6 months ────────────────────────────────────────────
three_months_ago = pd.Timestamp("2025-12-20")
recent_panel = df_icsa[df_icsa["date"] >= three_months_ago].copy()
ax_right.bar(recent_panel["date"], recent_panel["initial_claims"] / 1000,
width=5, color=COLORS['primary'], alpha=0.4)
ax_right.plot(recent_panel["date"], recent_panel["initial_claims_4wma"] / 1000,
color=COLORS['primary'], linewidth=2.5, zorder=5)
# Reference line at since-2023 mean
mean_val = stats['icsa_mean_since_2023'] / 1000
ax_right.axhline(y=mean_val, color=COLORS['neutral'], linestyle='--', linewidth=1, alpha=0.4)
ax_right.text(recent_panel["date"].iloc[0], mean_val + 1.5,
f'Avg since 2023: {mean_val:.0f}K', fontsize=7, color=COLORS['neutral'], alpha=0.7)
# Annotate latest
latest_row = recent_panel.iloc[-1]
ax_right.annotate(f"{latest_row['initial_claims']/1000:.0f}K",
xy=(latest_row["date"], latest_row["initial_claims"] / 1000),
xytext=(-15, 15), textcoords='offset points',
fontsize=10, fontweight='bold', color=COLORS['primary'],
arrowprops=dict(arrowstyle='->', color=COLORS['primary'], lw=1.5))
ax_right.set_ylabel("")
ax_right.set_ylim(top=300)
ax_right.xaxis.set_major_formatter(mdates.DateFormatter('%b\n%d'))
ax_right.xaxis.set_major_locator(mdates.MonthLocator())
ax_right.yaxis.set_major_formatter(ticker.FuncFormatter(lambda x, _: f"{x:.0f}K"))
ax_right.set_title("Last Three Months",
fontsize=12, fontweight='bold', pad=10)
fig.suptitle("Initial Unemployment Claims", fontsize=13, fontweight='bold', y=1.02)
plt.tight_layout()
Two features stand out. First, the left panel shows that the normalization from pandemic-era levels was essentially complete by mid-2023, and claims have traded in a remarkably narrow band since then. The 195,000-to-259,000 range over the past three years is tight by historical standards. Second, the right panel makes the recent dynamics visible: the late-January spike to 230,000 has fully reversed, and the 4-week MA is trending down through March.
If initial claims tell you how many people are losing jobs, continuing claims tell you how quickly they are finding new ones. A rising continuing claims number means displaced workers are staying on benefits longer, a sign of a labor market where re-employment is getting harder. The opposite is happening now.
# =============================================================================
# Continuing Claims: Full History with 4-Week Moving Average
# Source: data/clean/continuing_claims.csv (FRED series CCSA)
# =============================================================================
fig, ax = plt.subplots(figsize=(8, 5))
# Plot weekly continuing claims
ax.plot(df_ccsa["date"], df_ccsa["continuing_claims"] / 1e6,
color=COLORS['light'], linewidth=0.8, alpha=0.6, label='Weekly')
# Plot 4-week moving average
ax.plot(df_ccsa["date"], df_ccsa["continuing_claims_4wma"] / 1e6,
color=COLORS['accent'], linewidth=2.5, label='4-Week MA')
# Latest value annotation
latest_c_row = df_ccsa.dropna(subset=["continuing_claims_4wma"]).iloc[-1]
ax.annotate(f"{latest_c_row['continuing_claims']/1e6:.3f}M",
xy=(latest_c_row["date"], latest_c_row["continuing_claims"] / 1e6),
xytext=(15, 15), textcoords='offset points',
fontsize=10, fontweight='bold', color=COLORS['accent'],
arrowprops=dict(arrowstyle='->', color=COLORS['accent'], lw=1.5))
# Reference: since-2023 mean
mean_ccsa = stats['ccsa_mean_since_2023'] / 1e6
ax.axhline(y=mean_ccsa, color=COLORS['neutral'], linestyle='--', linewidth=1, alpha=0.4)
ax.text(df_ccsa["date"].iloc[5], mean_ccsa + 0.015,
f'Avg since 2023: {mean_ccsa:.2f}M', fontsize=8, color=COLORS['neutral'], alpha=0.7)
ax.set_ylabel("Continuing Claims (millions)")
ax.set_xlabel("")
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b\n%Y'))
ax.xaxis.set_major_locator(mdates.MonthLocator(interval=6))
ax.yaxis.set_major_formatter(ticker.FuncFormatter(lambda x, _: f"{x:.2f}M"))
ax.legend(loc='upper right', fontsize=9, framealpha=0.9)
ax.set_title("Continuing Claims: Workers Are Finding Jobs",
fontsize=13, fontweight='bold', pad=12)
plt.tight_layout()
Continuing claims dropped 32,000 to 1,819,000, the lowest reading since May 25, 2024. Year-over-year, continuing claims are down 57,000 from 1,876,000. The 4-week moving average at 1,847,000 has been drifting lower since the start of the year. This is the most underappreciated signal in the data: even as businesses report hiring caution in surveys, workers who do lose their jobs are cycling back into employment at a healthy pace.
The late-January/early-February spike in initial claims deserves closer examination because it overlapped with the survey reference period for the February employment report, the one that produced the payroll miss. Did the claims spike signal real weakness, or was it noise?
# =============================================================================
# The February Spike: Claims During the Payroll Survey Window
# Source: data/clean/initial_claims.csv (FRED series ICSA)
# =============================================================================
six_months_ago = pd.Timestamp("2025-09-20")
recent = df_icsa[df_icsa["date"] >= six_months_ago].copy()
fig, ax = plt.subplots(figsize=(8, 5))
# Weekly bars
ax.bar(recent["date"], recent["initial_claims"] / 1000,
width=5, color=COLORS['primary'], alpha=0.4)
# 4-week MA line
ax.plot(recent["date"], recent["initial_claims_4wma"] / 1000,
color=COLORS['primary'], linewidth=2.5, zorder=5)
# Highlight the Feb survey reference period (approx Jan 5 - Feb 1, 2026)
ax.axvspan(pd.Timestamp("2026-01-05"), pd.Timestamp("2026-02-01"),
alpha=0.08, color=COLORS['warning'],
label='Feb payroll survey window')
# Highlight the March survey reference period
ax.axvspan(pd.Timestamp("2026-02-08"), pd.Timestamp("2026-03-14"),
alpha=0.06, color=COLORS['positive'],
label='Mar payroll survey window')
# Annotate the spike
spike_date = pd.Timestamp("2026-01-31")
ax.annotate(f"230K\nFeb spike",
xy=(spike_date, 230),
xytext=(-50, 20), textcoords='offset points',
fontsize=9, color=COLORS['warning'], fontweight='bold',
arrowprops=dict(arrowstyle='->', color=COLORS['warning'], lw=1.5))
# Annotate the March reference week
mar_ref_date = pd.Timestamp("2026-03-14")
mar_ref_val = recent.loc[recent["date"] == mar_ref_date, "initial_claims"]
if len(mar_ref_val) > 0:
ax.annotate(f"{int(mar_ref_val.values[0]/1000)}K\nMar ref week",
xy=(mar_ref_date, mar_ref_val.values[0] / 1000),
xytext=(15, 25), textcoords='offset points',
fontsize=9, color=COLORS['positive'], fontweight='bold',
arrowprops=dict(arrowstyle='->', color=COLORS['positive'], lw=1.5))
# Reference line at since-2023 mean
mean_val = stats['icsa_mean_since_2023'] / 1000
ax.axhline(y=mean_val, color=COLORS['neutral'], linestyle='--', linewidth=1, alpha=0.4)
ax.text(recent["date"].iloc[1], mean_val + 1,
f'Avg since 2023: {mean_val:.0f}K', fontsize=8, color=COLORS['neutral'], alpha=0.7)
ax.set_ylabel("Initial Claims (thousands)")
ax.set_xlabel("")
ax.set_ylim(top=300)
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %d'))
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.yaxis.set_major_formatter(ticker.FuncFormatter(lambda x, _: f"{x:.0f}K"))
ax.legend(loc='upper right', fontsize=9, framealpha=0.9)
ax.set_title("Claims During Payroll Survey Windows",
fontsize=13, fontweight='bold', pad=12)
plt.tight_layout()
The answer is noise. The spike to 230,000 lasted exactly two weeks before claims resumed their downward drift. The March survey reference period (the week including March 12) saw initial claims at 205,000, one of the lowest readings of the year. If the February payroll miss was partly driven by the claims spike during its survey window, the March report should benefit from the opposite dynamic: claims falling during its reference period.
The relationship between initial and continuing claims reveals the flow of the labor market. When initial claims are stable but continuing claims are falling, it means the inflow of new layoffs is steady while the outflow (people finding new jobs) is accelerating. That is exactly what the data shows.
# =============================================================================
# Claims Ratio: Continuing / Initial (4-Week MAs)
# Source: data/clean/initial_claims.csv, continuing_claims.csv
# =============================================================================
# Merge on date (CCSA is lagged one week behind ICSA)
# Align by shifting CCSA dates forward one week for comparison
df_merged = pd.merge(
df_icsa[["date", "initial_claims_4wma"]].rename(columns={"date": "date"}),
df_ccsa[["date", "continuing_claims_4wma"]].assign(
date=df_ccsa["date"] + pd.Timedelta(days=7)
),
on="date", how="inner"
).dropna()
df_merged["ratio"] = df_merged["continuing_claims_4wma"] / df_merged["initial_claims_4wma"]
# Filter to since 2023
df_ratio = df_merged[df_merged["date"] >= "2023-01-01"].copy()
fig, ax = plt.subplots(figsize=(8, 5))
ax.plot(df_ratio["date"], df_ratio["ratio"],
color=COLORS['primary'], linewidth=2.5)
ax.fill_between(df_ratio["date"], df_ratio["ratio"],
df_ratio["ratio"].min(), alpha=0.08, color=COLORS['primary'])
# Latest annotation
last = df_ratio.iloc[-1]
ax.annotate(f"{last['ratio']:.1f}x",
xy=(last["date"], last["ratio"]),
xytext=(15, 10), textcoords='offset points',
fontsize=10, fontweight='bold', color=COLORS['primary'],
arrowprops=dict(arrowstyle='->', color=COLORS['primary'], lw=1.5))
# Mean line
mean_ratio = df_ratio["ratio"].mean()
ax.axhline(y=mean_ratio, color=COLORS['neutral'], linestyle='--', linewidth=1, alpha=0.4)
ax.text(df_ratio["date"].iloc[2], mean_ratio + 0.05,
f'Avg: {mean_ratio:.1f}x', fontsize=8, color=COLORS['neutral'], alpha=0.7)
ax.set_ylabel("Continuing / Initial Claims Ratio")
ax.set_xlabel("")
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b\n%Y'))
ax.xaxis.set_major_locator(mdates.MonthLocator(interval=3))
ax.set_title("Claims Duration Ratio: Re-employment Is Accelerating",
fontsize=13, fontweight='bold', pad=12)
plt.tight_layout()
The declining ratio is good news for the April 3 report. Even if net job creation was modest in March, the flow data suggests that the pool of unemployed workers is shrinking. People who lose jobs are not staying unemployed for long. This is inconsistent with the kind of labor market deterioration that would produce a second consecutive payroll miss.
The claims data heading into the April 3 jobs report is about as clean as it gets. Initial claims are low and stable, continuing claims are at a nearly two-year low, and the February spike that may have contaminated the payroll survey has fully reversed. The base case is a rebound in March payrolls.
For the Fed: Claims at 210,000 and an insured unemployment rate of 1.2% leave no room to argue that the labor market is deteriorating. The February payroll miss increasingly looks like an anomaly, not a trend break. If March payrolls rebound, the Fed’s case for holding rates steady strengthens further. Any rate action will be driven by the inflation side, not the labor side.
For workers: The continuing claims data is the most encouraging signal. At 1,819,000, displaced workers are finding re-employment faster than at any point since mid-2024. The risk is not layoffs but the hiring freeze documented in the BTOS data: fewer new openings even as existing jobs remain secure.
For markets: A strong March payroll number, supported by the claims trend, would reinforce the soft-landing narrative and likely push rate-cut expectations further out. A second miss, given the claims backdrop, would raise questions about structural shifts in the labor market that weekly data cannot capture.
| Series | Description | Source |
|---|---|---|
| ICSA | Initial Claims, Seasonally Adjusted (weekly) | FRED Series ICSA |
| CCSA | Continued Claims, Seasonally Adjusted (weekly) | FRED Series CCSA |
| DOL Weekly Report | Unemployment Insurance Weekly Claims | DOL News Release (March 26, 2026) |
Data current as of March 26, 2026. Claims data through week ending March 21, 2026. The April employment situation report is scheduled for April 3, 2026.