The U.S. Job Market: Openings, Quits, Layoffs, and What the Data Says

1. Motivation

The U.S. labor market has gone through dramatic swings over the past decade a historic tightening post-pandemic, record job openings in between 2021–2022, a surge in people voluntaringly quitting dubbed the “Great Resignation,” and then a gradual normalization through 2023–2024. As someone actively navigating this job market, I want to move beyond headlines and ask: what do the numbers actually show?

Specifically, this project investigates the following :

How job openings, quit rates, and unemployment have moved together over time
Whether macro labor indicators can statistically predict unemployment
What a short-term forecast of job openings looks like going into 2025–2026

This analysis follows the OSEMN workflow: Obtain → Scrub → Explore → Model → iNterpret.

2. Data Sources

This project draws from two distinct data source types:

FRED Public CSV Downloads (Federal Reserve Bank of St. Louis) — FRED provides direct public CSV downloads for every economic series at a stable URL, requiring no API key or account. This gives us clean monthly time-series data for four key labor market indicators.
BLS JOLTS Flat File (Bureau of Labor Statistics) — a raw tab delimited text file downloaded directly from the BLS public data server. Unlike the clean FRED CSVs, this file requires substantial cleaning and restructuring before it can be used.

3. Data Acquisition

3.1 Source 1: FRED Public CSV Downloads

FRED makes every data series available as a direct CSV download at a public URL — no account or API key required.

# Helper: download a FRED series by ID and return a clean two-column dataframe
# Uses download.file() to a temp file for broad HTTPS compatibility
load_fred <- function(series_id, col_name,
                      start = "2010-01-01", end = "2024-12-01") {
  url      <- paste0("https://fred.stlouisfed.org/graph/fredgraph.csv?id=", series_id)
  tmp      <- tempfile(fileext = ".csv")
  download.file(url, destfile = tmp, mode = "wb", quiet = TRUE)
  df       <- read_csv(tmp, col_types = cols(.default = col_character()),
                       show_col_types = FALSE)
  colnames(df) <- c("date", col_name)
  df %>%
    mutate(date      = as.Date(date),
           !!col_name := as.numeric(.data[[col_name]])) %>%
    filter(date >= as.Date(start), date <= as.Date(end))
}

# Pull four series
job_openings  <- load_fred("JTSJOL",   "job_openings_k")    # Job Openings (thousands)
unemployment  <- load_fred("UNRATE",   "unemployment_rate")  # Unemployment Rate (%)
quit_rate     <- load_fred("JTSQUR",   "quit_rate")          # Quits Rate (%)
fed_funds     <- load_fred("FEDFUNDS", "fed_funds_rate")     # Fed Funds Rate (%)

cat("Rows loaded per series:\n")

## Rows loaded per series:

cat("  Job Openings: ",  nrow(job_openings),  "\n")

##   Job Openings:  180

cat("  Unemployment: ",  nrow(unemployment),  "\n")

##   Unemployment:  180

cat("  Quit Rate:    ",  nrow(quit_rate),     "\n")

##   Quit Rate:     180

cat("  Fed Funds:    ",  nrow(fed_funds),     "\n")

##   Fed Funds:     180

3.2 Source 2: BLS JOLTS Flat File

The BLS publishes raw JOLTS data as a tab-delimited flat file at a public URL. Unlike the FRED CSVs, this file contains all JOLTS series in one file and requires filtering, parsing, and restructuring.

bls_url <- "https://download.bls.gov/pub/time.series/jt/jt.data.0.Current"

bls_raw <- tryCatch({
  tmp_bls <- tempfile(fileext = ".txt")
  download.file(bls_url, destfile = tmp_bls, mode = "wb", quiet = TRUE)
  read_tsv(tmp_bls, col_types = cols(.default = "c"), show_col_types = FALSE)
}, error = function(e) {
  message("BLS file could not be downloaded: ", e$message)
  NULL
})

if (!is.null(bls_raw)) {
  cat("BLS file loaded:", nrow(bls_raw), "rows,", ncol(bls_raw), "columns\n")
  cat("Column names:", paste(names(bls_raw), collapse = ", "), "\n")
} else {
  cat("BLS file unavailable — cross-validation section will be skipped.\n")
}

## BLS file unavailable — cross-validation section will be skipped.

4. Data Cleaning & Transformation

4.1 Merge and Transform FRED Data

# Join all four FRED series into one wide dataframe
fred_wide <- job_openings %>%
  inner_join(unemployment, by = "date") %>%
  inner_join(quit_rate,    by = "date") %>%
  inner_join(fed_funds,    by = "date") %>%
  mutate(
    job_openings_m = job_openings_k / 1000,  # Convert thousands -> millions
    year           = year(date),
    month          = month(date)
  ) %>%
  filter(!is.na(job_openings_m), !is.na(unemployment_rate),
         !is.na(quit_rate),      !is.na(fed_funds_rate))

# --- DATA TRANSFORMATION: Wide to Long ---
# Pivot to long format for multi-series time-series plotting
fred_long <- fred_wide %>%
  select(date, job_openings_m, unemployment_rate, quit_rate, fed_funds_rate) %>%
  pivot_longer(
    cols      = -date,
    names_to  = "indicator",
    values_to = "value"
  ) %>%
  mutate(indicator = recode(indicator,
    "job_openings_m"    = "Job Openings (millions)",
    "unemployment_rate" = "Unemployment Rate (%)",
    "quit_rate"         = "Quit Rate (%)",
    "fed_funds_rate"    = "Fed Funds Rate (%)"
  ))

head(fred_long, 8) %>%
  kable(caption = "Sample: Long-format FRED data after pivot_longer()") %>%
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Sample: Long-format FRED data after pivot_longer()
date	indicator	value
2010-01-01	Job Openings (millions)	2.837
2010-01-01	Unemployment Rate (%)	9.800
2010-01-01	Quit Rate (%)	1.300
2010-01-01	Fed Funds Rate (%)	0.110
2010-02-01	Job Openings (millions)	2.666
2010-02-01	Unemployment Rate (%)	9.800
2010-02-01	Quit Rate (%)	1.400
2010-02-01	Fed Funds Rate (%)	0.130

4.2 Clean BLS Flat File

The BLS flat file requires trimming whitespace, filtering to the total nonfarm series, removing annual averages (period M13), parsing period strings (M01-M12) into dates, and filtering to the study window.

if (!is.null(bls_raw)) {
  bls_clean <- bls_raw %>%
    mutate(across(where(is.character), str_trim)) %>%
    filter(str_detect(series_id, "JTS00000000000000")) %>%
    filter(period != "M13") %>%
    mutate(
      month_num = as.integer(str_remove(period, "M")),
      year_num  = as.integer(year),
      date      = make_date(year_num, month_num, 1L),
      value     = as.numeric(value)
    ) %>%
    filter(!is.na(date),
           date >= as.Date("2010-01-01"),
           date <= as.Date("2024-12-01")) %>%
    select(series_id, date, value) %>%
    arrange(date)

  cat("BLS series present after cleaning:\n")
  bls_clean %>% count(series_id) %>% print()
} else {
  bls_clean <- NULL
}

4.3 Summary Statistics

fred_wide %>%
  select(job_openings_m, unemployment_rate, quit_rate, fed_funds_rate) %>%
  summary() %>%
  kable(caption = "Summary Statistics: Key Labor Market Indicators (2010-2024)") %>%
  kable_styling(bootstrap_options = c("striped", "hover"))

Summary Statistics: Key Labor Market Indicators (2010-2024)
job_openings_m	unemployment_rate	quit_rate	fed_funds_rate
Min. : 2.666	Min. : 3.400	Min. :1.300	Min. :0.050
1st Qu.: 4.128	1st Qu.: 3.900	1st Qu.:1.700	1st Qu.:0.090
Median : 6.074	Median : 5.000	Median :2.100	Median :0.190
Mean : 6.356	Mean : 5.797	Mean :2.048	Mean :1.229
3rd Qu.: 7.442	3rd Qu.: 7.525	3rd Qu.:2.300	3rd Qu.:1.823
Max. :12.301	Max. :14.800	Max. :3.000	Max. :5.330

5. Exploratory Data Analysis

5.1 All Indicators Over Time

recessions <- data.frame(
  start = as.Date("2020-02-01"),
  end   = as.Date("2020-04-01")
)

ggplot(fred_long, aes(x = date, y = value, color = indicator)) +
  geom_rect(data = recessions,
            aes(xmin = start, xmax = end, ymin = -Inf, ymax = Inf),
            inherit.aes = FALSE, fill = "grey80", alpha = 0.5) +
  geom_line(linewidth = 0.8) +
  facet_wrap(~ indicator, scales = "free_y", ncol = 2) +
  scale_x_date(date_breaks = "2 years", date_labels = "%Y") +
  scale_color_brewer(palette = "Set1") +
  labs(
    title    = "U.S. Labor Market Indicators, 2010-2024",
    subtitle = "Shaded area = COVID-19 recession (Feb-Apr 2020)",
    x = NULL, y = "Value", color = NULL,
    caption  = "Source: FRED (fred.stlouisfed.org)"
  ) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "none",
        strip.text = element_text(face = "bold"))

5.2 The Great Resignation in Focus (2019-2023)

fred_wide %>%
  filter(date >= as.Date("2019-01-01"), date <= as.Date("2023-12-01")) %>%
  select(date, job_openings_m, quit_rate) %>%
  pivot_longer(-date, names_to = "indicator", values_to = "value") %>%
  mutate(indicator = recode(indicator,
    "job_openings_m" = "Job Openings (millions)",
    "quit_rate"      = "Quit Rate (%)"
  )) %>%
  ggplot(aes(x = date, y = value, color = indicator)) +
  geom_line(linewidth = 1.1) +
  annotate("rect",
           xmin = as.Date("2021-03-01"), xmax = as.Date("2022-06-01"),
           ymin = -Inf, ymax = Inf, fill = "lightyellow", alpha = 0.4) +
  annotate("text", x = as.Date("2021-10-01"), y = 11.8,
           label = "\"Great Resignation\"\nPeak", size = 3.5, color = "darkorange") +
  facet_wrap(~ indicator, scales = "free_y") +
  scale_x_date(date_breaks = "6 months", date_labels = "%b %Y") +
  scale_color_manual(values = c("#E41A1C", "#377EB8")) +
  labs(
    title    = "Job Openings and Quit Rate: The Great Resignation Era",
    subtitle = "Both indicators peaked in 2021-2022 before declining sharply",
    x = NULL, y = NULL, color = NULL,
    caption  = "Source: FRED"
  ) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "none",
        axis.text.x = element_text(angle = 45, hjust = 1),
        strip.text  = element_text(face = "bold"))

5.3 Beveridge Curve (Job Openings vs. Unemployment)

The Beveridge Curve plots job openings against unemployment over time. A healthy economy shows an inverse relationship; outward shifts suggest structural labor market mismatch between available workers and available jobs.

fred_wide %>%
  mutate(era = case_when(
    date < as.Date("2020-01-01")                                 ~ "Pre-COVID (2010-2019)",
    date >= as.Date("2020-01-01") & date < as.Date("2021-07-01") ~ "COVID Shock (2020-mid 2021)",
    date >= as.Date("2021-07-01") & date < as.Date("2023-01-01") ~ "Great Resignation (2021-2022)",
    TRUE                                                          ~ "Normalization (2023-2024)"
  )) %>%
  ggplot(aes(x = unemployment_rate, y = job_openings_m, color = era)) +
  geom_path(aes(group = 1), color = "grey70", linewidth = 0.4) +
  geom_point(size = 2, alpha = 0.8) +
  scale_color_brewer(palette = "Set1") +
  labs(
    title    = "Beveridge Curve: Job Openings vs. Unemployment Rate (2010-2024)",
    subtitle = "Each point = one month. Path shows trajectory over time.",
    x        = "Unemployment Rate (%)",
    y        = "Job Openings (millions)",
    color    = "Era",
    caption  = "Source: FRED"
  ) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "bottom")

5.4 Job Openings Distribution by Era

fred_wide %>%
  mutate(era = case_when(
    date < as.Date("2020-01-01")  ~ "Pre-COVID",
    date < as.Date("2021-07-01")  ~ "COVID Shock",
    date < as.Date("2023-01-01")  ~ "Great Resignation",
    TRUE                          ~ "Normalization"
  )) %>%
  mutate(era = factor(era, levels = c("Pre-COVID", "COVID Shock",
                                      "Great Resignation", "Normalization"))) %>%
  ggplot(aes(x = era, y = job_openings_m, fill = era)) +
  geom_boxplot(alpha = 0.7, outlier.shape = 21) +
  scale_fill_brewer(palette = "Set2") +
  labs(
    title   = "Distribution of Job Openings by Labor Market Era",
    x       = NULL,
    y       = "Job Openings (millions)",
    fill    = NULL,
    caption = "Source: FRED"
  ) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "none")

6. Statistical Analysis

6.1 Correlation Matrix

cor_data <- fred_wide %>%
  select(job_openings_m, unemployment_rate, quit_rate, fed_funds_rate) %>%
  drop_na()

cor_matrix <- cor(cor_data, method = "pearson")

corrplot(
  cor_matrix,
  method      = "color",
  type        = "upper",
  addCoef.col = "black",
  tl.col      = "black",
  tl.srt      = 45,
  col         = colorRampPalette(c("#E41A1C", "white", "#377EB8"))(200),
  title       = "Pearson Correlation Matrix: Labor Market Indicators (2010-2024)",
  mar         = c(0, 0, 2, 0)
)

Key findings:

Job openings and quit rate are strongly positively correlated — when employers compete aggressively for workers, employees feel confident enough to quit voluntarily.
Job openings and unemployment are strongly negatively correlated — this is the Beveridge Curve relationship expressed as a single number.
Fed funds rate and unemployment are negatively correlated — rate hikes tend to follow tight labor markets.

6.2 Multiple Linear Regression: Predicting Unemployment

Can job openings, quit rate, and the federal funds rate together statistically predict the unemployment rate?

model <- lm(unemployment_rate ~ job_openings_m + quit_rate + fed_funds_rate,
            data = fred_wide)

tidy(model) %>%
  mutate(across(where(is.numeric), ~ round(.x, 4))) %>%
  kable(caption = "Regression Coefficients: Predictors of Unemployment Rate") %>%
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Regression Coefficients: Predictors of Unemployment Rate
term	estimate	std.error	statistic
(Intercept)	16.8487	0.7592	22.1929
job_openings_m	0.7368	0.1329	5.5454
quit_rate	-7.2996	0.7256	-10.0605
fed_funds_rate	-0.6396	0.0639	-10.0040

glance(model) %>%
  select(r.squared, adj.r.squared, p.value, AIC) %>%
  mutate(across(everything(), ~ round(.x, 4))) %>%
  kable(caption = "Model Fit Statistics") %>%
  kable_styling(bootstrap_options = c("striped", "hover"), full_width = FALSE)

Model Fit Statistics
r.squared	adj.r.squared	p.value	AIC
0.7314	0.7268	0	573.1738

fred_wide_model <- fred_wide %>%
  drop_na(job_openings_m, quit_rate, fed_funds_rate) %>%
  mutate(.fitted = fitted(model))

ggplot(fred_wide_model, aes(x = date)) +
  geom_line(aes(y = unemployment_rate, color = "Actual"),  linewidth = 0.9) +
  geom_line(aes(y = .fitted,           color = "Fitted"),  linewidth = 0.9, linetype = "dashed") +
  scale_color_manual(values = c("Actual" = "#E41A1C", "Fitted" = "#377EB8")) +
  labs(
    title    = "Actual vs. Fitted Unemployment Rate",
    subtitle = paste0("Adj. R2 = ", round(summary(model)$adj.r.squared, 3),
                      " — Model explains most variance outside the COVID shock"),
    x = NULL, y = "Unemployment Rate (%)", color = NULL,
    caption = "Source: FRED"
  ) +
  theme_minimal(base_size = 12) +
  theme(legend.position = "top")

Interpretation: The model achieves a high adjusted R², meaning these three indicators together explain the vast majority of month-to-month variation in unemployment. The largest residuals occur during the COVID shock in March–April 2020 — expected, since that spike was driven by an unprecedented external event rather than gradual labor market dynamics.

7. Beyond-Class Feature: Time-Series Forecasting with Prophet

prophet is a forecasting library developed by Meta (Facebook) that was not covered in coursework. It automatically detects trend changepoints, handles yearly seasonality, and produces uncertainty intervals — well-suited to labor market time series with structural breaks like COVID.

# Prophet requires columns named 'ds' (date) and 'y' (value)
prophet_df <- fred_wide %>%
  select(ds = date, y = job_openings_m) %>%
  drop_na()

# Fit model with higher changepoint flexibility to capture the post-COVID structural break
m <- prophet(
  prophet_df,
  yearly.seasonality      = TRUE,
  weekly.seasonality      = FALSE,
  daily.seasonality       = FALSE,
  changepoint.prior.scale = 0.3
)

# Forecast 18 months ahead
future      <- make_future_dataframe(m, periods = 18, freq = "month")
forecast_df <- predict(m, future)

# Plot
plot(m, forecast_df) +
  labs(
    title    = "Prophet Forecast: U.S. Job Openings (millions)",
    subtitle = "Historical data through 2024; shaded band = 80% uncertainty interval",
    x = NULL, y = "Job Openings (millions)",
    caption  = "Source: FRED | Forecast: Meta Prophet"
  ) +
  theme_minimal(base_size = 12)

prophet_plot_components(m, forecast_df)

Interpretation: Prophet identifies a clear structural upward shift in job openings post-2020 and a subsequent normalization. The seasonal component reveals a mild annual cycle — openings tend to dip slightly in late winter and pick up in spring. The 18-month forecast projects continued stabilization around 7-8 million openings, consistent with a gradually softening but still historically tight labor market.

8. BLS Cross-Source Validation

We use the BLS flat file to cross-validate FRED’s total nonfarm job openings figures. Agreement between independent sources confirms data integrity.

if (!is.null(bls_clean)) {
  bls_jol <- bls_clean %>%
    filter(str_detect(series_id, "JOL")) %>%
    mutate(bls_openings_m = value / 1000) %>%
    select(date, bls_openings_m)

  comparison <- fred_wide %>%
    select(date, job_openings_m) %>%
    inner_join(bls_jol, by = "date") %>%
    drop_na()

  comparison %>%
    summarise(
      n            = n(),
      mean_diff    = round(mean(job_openings_m - bls_openings_m, na.rm = TRUE), 5),
      max_abs_diff = round(max(abs(job_openings_m - bls_openings_m), na.rm = TRUE), 5),
      correlation  = round(cor(job_openings_m, bls_openings_m, use = "complete.obs"), 6)
    ) %>%
    kable(caption = "FRED vs. BLS Cross-Source Validation") %>%
    kable_styling(bootstrap_options = c("striped"), full_width = FALSE)

  comparison %>%
    pivot_longer(-date, names_to = "source", values_to = "openings") %>%
    mutate(source = recode(source,
      "job_openings_m"  = "FRED CSV",
      "bls_openings_m"  = "BLS Flat File"
    )) %>%
    ggplot(aes(x = date, y = openings, color = source, linetype = source)) +
    geom_line(linewidth = 0.9) +
    scale_color_manual(values = c("FRED CSV" = "#377EB8", "BLS Flat File" = "#E41A1C")) +
    labs(
      title    = "Cross-Source Validation: FRED vs. BLS Job Openings",
      subtitle = "Series should track near-identically — any gap indicates a reporting lag",
      x = NULL, y = "Job Openings (millions)",
      color = "Source", linetype = "Source",
      caption = "Sources: FRED; BLS JOLTS flat file"
    ) +
    theme_minimal(base_size = 12) +
    theme(legend.position = "top")
} else {
  cat("BLS cross-validation skipped — flat file was unavailable at time of knitting.\n")
}

## BLS cross-validation skipped — flat file was unavailable at time of knitting.

9. Conclusions

This project analyzed U.S. labor market dynamics from 2010 through 2024 using two data source types (FRED public CSV and BLS flat file) and the OSEMN workflow.