── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
✔ dplyr 1.1.4 ✔ readr 2.1.6
✔ forcats 1.0.1 ✔ stringr 1.6.0
✔ ggplot2 4.0.1 ✔ tibble 3.3.0
✔ lubridate 1.9.4 ✔ tidyr 1.3.2
✔ purrr 1.1.0
── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()
ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
Attaching package: 'scales'
The following object is masked from 'package:purrr':
discard
The following object is masked from 'package:readr':
col_factorU.S. Weekly Retail Fuel Prices: Seasonality and the Diesel Premium
Rows: 22360 Columns: 5
── Column specification ────────────────────────────────────────────────────────
Delimiter: ","
chr (3): fuel, grade, formulation
dbl (1): price
date (1): date
ℹ Use `spec()` to retrieve the full column specification for this data.
ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.Introduction
The dataset we analyze comes from the U.S. Energy Information Administration (EIA) and reports weekly national average retail fuel prices in the United States. Prices are collected every Monday from a large sample of gasoline stations and diesel outlets across the country and represent the average cash pump price paid by consumers, including all taxes. The data include both gasoline and diesel prices (USD per gallon) and, for gasoline, breakouts by grade (regular, midgrade, premium, and all grades) and formulation (conventional vs. reformulated). Each row corresponds to a specific fuel configuration in a given week.
The dataset contains 22360 observations spanning 1813 weeks from 1990-08-20 to 2025-06-23. This weekly frequency and multi-decade span make the data well suited for examining both seasonal patterns and longer-run shifts in fuel markets, while still capturing short-term shocks. In this analysis, we focus on (1) whether seasonality differs meaningfully across fuel type, gasoline grade, and gasoline formulation, and (2) how the diesel premium (diesel minus all-grades gasoline) evolves over time, particularly during periods of higher versus lower price volatility.
So What? The results of this analysis are crucial for understanding the dynamics of fuel prices across different fuel types, grades, and formulations over time. Insights into seasonality patterns can help stakeholders in the fuel industry (policymakers, fuel suppliers, and consumers) for price fluctuations tied to seasonal demand changes. Additionally, analyzing the diesel premium’s evolution during periods of volatility can reveal critical information for industries reliant on diesel fuel, such as transportation and logistics. Understanding these trends is vital for making informed decisions on pricing, economic forecasting, and managing supply chain costs, ultimately influencing both policy and business strategies in the energy sector.
Question 1 <- How Do Weekly U.S. Retail Fuel Prices Vary by Season Across Fuel Type and Gasoline Grade?
Introduction
Seasonal patterns in fuel prices are a well‑documented feature of U.S. energy markets. Gasoline demand typically rises during the summer driving season, while winter blends and regulatory requirements can influence prices at other times of year. Diesel prices, by contrast, are shaped more by freight activity, heating demand, and refinery constraints due to their different usages. Understanding whether these two fuels follow the same or different seasonal cycles matters because businesses and consumers who rely on each fuel face different cost pressures at different times of year. In this section, we examine how seasonal price patterns differ across two key dimensions: fuel type (gasoline vs. diesel) and gasoline grade (regular, midgrade, premium).
To do this, we use the weekly national average retail fuel price dataset from the U.S. Energy Information Administration. From the date variable, we create year, month, and a four‑category season variable (Winter, Spring, Summer, Fall). These derived variables allow us to compare how average prices vary across seasons and whether gasoline grades exhibit distinct seasonal behavior relative to each other and to diesel.
Approach
To examine seasonal differences across fuel types, we focus on weekly national average prices where each fuel has a single comparable observation per week. Because gasoline and diesel are structured differently in the dataset, gasoline observations are restricted to rows where grade == “all” and formulation == “all”, while diesel observations use rows where grade == “all”. We then compute mean prices within each season for gasoline and diesel separately. A line plot is used to emphasize directional seasonal change and allow direct comparison of how strongly each fuel responds to seasonal demand cycles.
To compare seasonal behavior across gasoline grades, we isolate gasoline observations where formulation == “all” and exclude aggregate “all” grade rows to avoid duplication. We visualize seasonal price distributions using boxplots, which highlight median prices, variability, and consistency across grades. Together, these visualizations allow us to distinguish between differences in seasonal magnitude across fuel types and stability of price hierarchy across gasoline grades, linking both figures into a single analytical narrative about seasonal fuel price dynamics.
Analysis
seasonal_data <- weekly_gas_prices |>
mutate(
month = month(date),
season = case_when(
month %in% c(12, 1, 2) ~ "Winter",
month %in% c(3, 4, 5) ~ "Spring",
month %in% c(6, 7, 8) ~ "Summer",
TRUE ~ "Fall"
),
# Order the seasons chronologically for easier plotting
season = factor(season, levels = c("Winter", "Spring", "Summer", "Fall"))
)fuel_seasonal_avg <- seasonal_data |>
filter(
(fuel == "gasoline" & grade == "all" & formulation == "all") |
(fuel == "diesel" & grade == "all")
) |>
group_by(season, fuel) |>
summarize(mean_price = mean(price, na.rm = TRUE), .groups = "drop")
# connecting points across seasons makes the trajectory easier to follow
# than a bar chart.
ggplot(
fuel_seasonal_avg,
aes(x = season, y = mean_price, group = fuel, color = fuel)
) +
geom_line(linewidth = 1.4) +
geom_point(size = 3) +
geom_text(
data = fuel_seasonal_avg |> filter(season == "Summer"),
aes(label = "Summer peak"),
vjust = -1,
show.legend = FALSE
) +
scale_y_continuous(
labels = dollar_format(),
expand = expansion(mult = c(0.05, 0.15))
) +
labs(
title = stringr::str_wrap(
"Gasoline Prices Exhibit Stronger Summer Seasonality Than Diesel",
width = 55
),
subtitle = "Weekly U.S. retail fuel prices averaged by season",
x = "Season",
y = "Average Price (USD per gallon)",
color = "Fuel Type",
caption = "Gasoline prices rise sharply during the summer driving season,
while diesel prices change more gradually across the year."
) +
theme_custom(base_size = 13)
gasoline_grade_data <- seasonal_data |>
filter(
fuel == "gasoline",
formulation == "all",
grade %in% c("regular", "midgrade", "premium")
) |>
mutate(
grade = factor(
grade,
levels = c("regular", "midgrade", "premium")
)
)
ggplot(
gasoline_grade_data,
aes(x = season, y = price, fill = grade)
) +
geom_boxplot(alpha = 0.85) +
stat_summary(
fun = median,
geom = "point",
shape = 18,
size = 3,
color = "black"
) +
# Dollar formatting improves readability, adding title and cross-chart interpretation
scale_y_continuous(labels = dollar_format()) +
labs(
title = stringr::str_wrap(
"Seasonal Changes Affect All Grades While Price Hierarchy Remains Stable",
width = 55
),
subtitle = "Distribution of weekly gasoline prices by season and grade",
x = "Season",
y = "Price (USD per gallon)",
fill = "Gasoline Grade",
caption = stringr::str_wrap(
"Although prices increase in summer across all grades, premium gasoline consistently remains the most expensive.",
width = 80
)
) +
theme_custom(base_size = 13)
Discussion
Together, the visualizations reveal that seasonal forces influence fuel markets differently depending on fuel type but consistently across gasoline grades. Gasoline prices display pronounced seasonality, rising sharply from winter to summer before declining in fall. This pattern reflects increased consumer travel demand and the higher production costs associated with environmentally regulated summer fuel blends. Diesel prices, although consistently higher overall, change more gradually across seasons and reach their highest levels later in the year, likely reflecting freight demand and heating-related fuel consumption rather than consumer travel behavior.
Across gasoline grades, seasonal movements occur simultaneously, but the relative ordering of prices remains unchanged. Premium gasoline remains consistently more expensive than midgrade and regular gasoline in every season, indicating that grade-based price differences are structurally maintained rather than driven by seasonal shocks. While seasonal demand raises prices across all grades, it primarily affects overall price levels rather than relative differences between fuel types.
Taken together, these findings suggest that seasonality primarily alters the magnitude of fuel prices rather than their relative structure. Gasoline markets appear more sensitive to seasonal consumer demand cycles, whereas diesel pricing reflects steadier commercial demand. This distinction highlights how different economic uses of fuels shape their exposure to seasonal market pressures.
Question 2 <- How Does the Diesel Premium Change Over Time, and How Is It Related to Price Volatility?
Introduction
The diesel premium (the difference between the weekly retail price of diesel and all-grades gasoline) is not a fixed margin. It narrows and widens as macroeconomic shocks, supply disruptions, and changes in freight demand ripple through energy markets. Understanding when and why this premium shifts can reveal how differently structured fuel markets respond to the same external pressures. We are interested in this question because diesel underpins virtually all goods transportation in the United States, so changes in its relative price carry broad economic implications beyond the pump.
To answer this question we use two variables derived from the EIA weekly price data: the diesel premium itself (diesel price minus all-grades average gasoline price) and a 12-week rolling standard deviation of weekly price changes, which we use as a measure of short-term price volatility. We classify each week into a Low, Medium, or High volatility regime using terciles of the diesel volatility distribution, and we examine how the premium behaves across those regimes and over calendar time.
Approach
Our first plot is a time-series line chart of the diesel premium with background shading indicating the volatility regime active in each week. A time-series line is the natural choice for showing how a continuous quantity evolves over decades, and layering the volatility shading directly behind the line allows the reader to see the regime context for every movement without needing to switch between panels. Vertical dashed lines mark the 2008 Financial Crisis and the COVID-19 onset, two well-known demand shocks that visibly affect both the premium and volatility classifications.
Our second plot is a scatter plot of diesel volatility against gasoline volatility, with each point representing one week. A scatter plot is ideal here because we want to understand the joint distribution of the two volatility measures — specifically, whether diesel tends to be more volatile than gasoline (points above the 45° reference line), and whether that relationship holds across all volatility levels. Points are colored by volatility regime so readers can see where the High-regime weeks cluster relative to the overall trend. A linear regression line with a confidence band summarizes the overall relationship, and density contours show where the data are most concentrated.
Analysis
weekly_premium <- weekly_gas_prices |>
filter(
fuel == "diesel" |
(fuel == "gasoline" & grade == "all")
) |>
group_by(date, fuel) |>
summarise(price_all = mean(price, na.rm = TRUE), .groups = "drop") |>
pivot_wider(names_from = fuel, values_from = price_all) |>
# lag() requires chronological order to produce correct week-over-week differences
arrange(date) |>
rename(
diesel_all = diesel,
gasoline_all_all = gasoline
) |>
mutate(
diesel_premium = diesel_all - gasoline_all_all,
gas_weekly_change = gasoline_all_all - lag(gasoline_all_all),
diesel_weekly_change = diesel_all - lag(diesel_all),
gas_vol_12w = slide_dbl(
gas_weekly_change,
sd,
.before = 11,
.complete = TRUE,
na.rm = TRUE
),
diesel_vol_12w = slide_dbl(
diesel_weekly_change,
sd,
.before = 11,
.complete = TRUE,
na.rm = TRUE
)
) |>
mutate(
vol_regime = case_when(
is.na(diesel_vol_12w) ~ NA_character_,
diesel_vol_12w <= quantile(diesel_vol_12w, 1 / 3, na.rm = TRUE) ~ "Low",
diesel_vol_12w <=
quantile(diesel_vol_12w, 2 / 3, na.rm = TRUE) ~ "Medium",
TRUE ~ "High"
),
# Explicit factor order ensures legend and axes read Low → Medium → High, not alphabetically.
vol_regime = factor(vol_regime, levels = c("Low", "Medium", "High"))
)crisis_dates <- tibble(
date = as.Date(c("2008-09-15", "2020-03-11")),
label = c("2008 Financial\nCrisis", "COVID-19\nOnset")
)
weekly_premium |>
filter(!is.na(diesel_premium), !is.na(vol_regime)) |>
arrange(date) |>
ggplot(aes(x = date, y = diesel_premium)) +
geom_rect(
aes(
xmin = date,
xmax = lead(date),
ymin = -Inf,
ymax = Inf,
fill = vol_regime
),
alpha = 0.15,
inherit.aes = FALSE
) +
geom_line(linewidth = 0.6) +
geom_vline(data = crisis_dates, aes(xintercept = date), linetype = "dashed") +
scale_y_continuous(
labels = dollar_format(prefix = "$"),
breaks = pretty_breaks(n = 6)
) +
scale_x_date(
date_breaks = "5 years",
date_labels = "%Y",
expand = expansion(mult = c(0.01, 0.02))
) +
labs(
title = "Diesel Premium Over Time (Diesel − All-Grades Gasoline)",
subtitle = "Background shading shows diesel price volatility regime (12-week rolling SD of weekly diesel changes)",
x = "Year",
y = "Diesel premium ($/gal)",
fill = "Diesel\nvolatility"
) +
theme_custom(base_size = 13) +
coord_cartesian(clip = "off")
weekly_data_filtered_navol <- weekly_premium |>
filter(!is.na(gas_vol_12w), !is.na(diesel_vol_12w))
# Shared axis limits force a square plot so the 45° reference line is a true diagonal.
max_val <- max(
weekly_data_filtered_navol$gas_vol_12w,
weekly_data_filtered_navol$diesel_vol_12w,
na.rm = TRUE
)
brks <- pretty_breaks(n = 6)(c(0, max_val))
weekly_data_filtered_navol |>
ggplot(aes(x = gas_vol_12w, y = diesel_vol_12w, color = vol_regime)) +
geom_point(alpha = 0.5, size = 1.6) +
geom_density_2d(color = "grey55", linewidth = 0.35) +
geom_smooth(
method = "lm",
se = TRUE,
color = "grey20",
fill = "grey75",
linewidth = 0.9
) +
# Points above the 45° line mean diesel is more volatile than gasoline that week.
geom_abline(slope = 1, intercept = 0, linetype = "dashed", color = "grey40") +
scale_x_continuous(
limits = c(0, max_val),
breaks = brks,
labels = dollar_format(prefix = "$"),
expand = c(0, 0)
) +
scale_y_continuous(
limits = c(0, max_val),
breaks = brks,
labels = dollar_format(prefix = "$"),
expand = c(0, 0)
) +
coord_equal() +
labs(
title = "Diesel vs. Gasoline Volatility",
subtitle = "Each point is one week; dashed line is the 45° equality reference",
x = "Gasoline Volatility (12-week SD, $/gal)",
y = "Diesel Volatility (12-week SD, $/gal)",
color = "Diesel\nVolatility"
) +
guides(color = guide_legend(override.aes = list(alpha = 1, size = 3))) +
theme_custom(base_size = 13)`geom_smooth()` using formula = 'y ~ x'
Discussion
The time-series chart shows that the diesel premium has been far from stable over the past three decades. For most of the 1990s and early 2000s, diesel traded at or below the price of all-grades gasoline, so the premium was near zero or slightly negative. Beginning around 2004, the premium shifted upward and became persistently positive, likely reflecting tighter global distillate markets and rising freight demand. The 2008 Financial Crisis coincides with a sharp spike followed by a rapid collapse in the premium — both events occurring within a High volatility regime — illustrating how sudden demand shocks can swing the premium dramatically. After COVID-19 onset in 2020, the premium initially fell as freight activity collapsed, then surged to its highest levels on record in 2022, again during a High volatility period.
The scatter plot reveals that diesel and gasoline volatility are strongly correlated — the regression line runs steeply upward and nearly all points fall above the 45° dashed reference line, indicating that diesel is systematically more volatile than gasoline across all levels of market stress. High-regime weeks (blue) cluster in the upper-right corner, confirming that both fuels become more volatile simultaneously during market disruptions, but diesel amplifies those swings. The widening vertical spread at higher gasoline volatility levels suggests that diesel’s response to market stress becomes increasingly unpredictable during the most turbulent periods.
Taken together, the two plots show that the diesel premium is a sensitive indicator of energy market stress. When volatility is high, the premium tends to be both larger and more unstable, amplifying cost pressures on freight-dependent industries. The persistent structural shift after 2004 and the outsized spikes during macroeconomic crises suggest that the diesel market is fundamentally less buffered than the gasoline market, making it a useful early signal for broader energy price instability.
Presentation
Our presentation can be found here.
Data
U.S. Energy Information Administration (EIA). (1990–present). Weekly US Gas Prices [Data set]. U.S. Department of Energy. Retrieved February 10, 2026, from https://www.eia.gov/dnav/pet/pet_pri_gnd_dcus_nus_w.htm. Data curated for TidyTuesday by Jon Harmon.
References
U.S. Energy Information Administration. (n.d.). Gasoline and diesel fuel update. U.S. Department of Energy. https://www.eia.gov/petroleum/gasdiesel/
R Core Team. (2023). stats::lag — Lag a time series. R Documentation. https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/lag
Davis, D. (2021). slider: Sliding window functions. R Documentation. https://www.rdocumentation.org/packages/slider/versions/0.3.3
R Core Team. (2023). stats::quantile — Sample quantiles. R Documentation. https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/quantile
Wickham, H. (2022). scales: Scale functions for visualization. R Documentation. https://rdrr.io/cran/scales/man/scales-package.html
user3680613. (2022). Change slide background using CSS with Quarto Reveal.js [Stack Overflow post]. Stack Overflow. https://stackoverflow.com/questions/74401855/change-slide-background-using-css-with-quarto-reveal-js